From 52ad8b232938c346a5a5a5589198207792ae2f49 Mon Sep 17 00:00:00 2001
From: Annalena Kofler <annalena.kofler@tuebingen.mpg.de>
Date: Wed, 22 Oct 2025 14:19:37 +0200
Subject: [PATCH 01/15] Fix version check for models that don't contain the
 version key (before v0.3.3). Important for old Zenodo models.

---
 dingo/core/posterior_models/build_model.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/dingo/core/posterior_models/build_model.py b/dingo/core/posterior_models/build_model.py
index ef4190120..1c38a0fae 100644
--- a/dingo/core/posterior_models/build_model.py
+++ b/dingo/core/posterior_models/build_model.py
@@ -44,7 +44,13 @@ def build_model_from_kwargs(
 
     if filename is not None:
         d, _ = torch_load_with_fallback(filename, preferred_map_location="meta")
-        check_minimum_version(d["version"])
+        if "version" in d:
+            check_minimum_version(d["version"])
+        else:
+            print(
+                f"Warning: The DINGO model loaded from {filename} was trained with a version older than v0.3.3. "
+            )
+            check_minimum_version("")
         update_model_config(d["metadata"]["train_settings"]["model"])  # Backward compat
         posterior_model_type = d["metadata"]["train_settings"]["model"][
             "posterior_model_type"

From 4b5aeb0784d87f594f6672fe6c719e0f7f2b4dff Mon Sep 17 00:00:00 2001
From: Annalena Kofler <annalena.kofler@tuebingen.mpg.de>
Date: Wed, 22 Oct 2025 14:24:28 +0200
Subject: [PATCH 02/15] Improve logging

---
 dingo/core/posterior_models/build_model.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/dingo/core/posterior_models/build_model.py b/dingo/core/posterior_models/build_model.py
index 1c38a0fae..2f7571032 100644
--- a/dingo/core/posterior_models/build_model.py
+++ b/dingo/core/posterior_models/build_model.py
@@ -47,10 +47,8 @@ def build_model_from_kwargs(
         if "version" in d:
             check_minimum_version(d["version"])
         else:
-            print(
-                f"Warning: The DINGO model loaded from {filename} was trained with a version older than v0.3.3. "
-            )
-            check_minimum_version("")
+            # version was introduced in v0.3.3
+            check_minimum_version("dingo=0.3.2")
         update_model_config(d["metadata"]["train_settings"]["model"])  # Backward compat
         posterior_model_type = d["metadata"]["train_settings"]["model"][
             "posterior_model_type"

From 174f913a480925db3a88743feede80a18d0b75c9 Mon Sep 17 00:00:00 2001
From: Heinrich Campe <hcampe@gmx.net>
Date: Tue, 30 Jun 2026 13:34:35 +0200
Subject: [PATCH 03/15] claude setup

---
 .gitignore |  4 ++++
 CLAUDE.md  | 39 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 43 insertions(+)
 create mode 100644 CLAUDE.md

diff --git a/.gitignore b/.gitignore
index 3a5bb4048..a6c7293c0 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,7 @@
+# codex stuff
+AGENTS*.md
+
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]
diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 000000000..81e9e3a34
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,39 @@
+Dingo: gravitational-wave parameter inference using neural posterior estimation.
+
+## Architecture
+- **`dingo/core/`** — model-agnostic ML/inference: `posterior_models/` (normalizing flows, flow matching, score diffusion), `nn/` (embedding nets, neural spline flows), `dataset.py` (HDF5 serialization), `result.py` (posterior samples + importance sampling).
+- **`dingo/gw/`** — GW-specific: `likelihood.py`, `domains/`, `waveform_generator/`, `transforms/`, `training/`, `inference/`.
+- **`dingo/pipe/`** — CLI pipeline (bilby_pipe-based): data generation → sampling → importance sampling → plotting. `parser.py` (INI), `main.py` (completed config + DAG), `nodes/` (HTCondor).
+- **`dingo/asimov/`** — Asimov (LVK) integration.
+
+## Commands
+- `uv sync`; `uv run pytest [tests/...::test]`; `uv run black dingo tests`.
+
+## Code writing — stop at the first option that solves the task correctly
+1. Don't build behavior that isn't needed.
+2. Use the standard library when it already solves the problem.
+3. Use an existing repository or platform feature.
+4. Use an already-installed dependency.
+5. Inline a one-line solution when it stays readable; don't wrap it in a one-line helper.
+6. Otherwise, write the minimum task-specific code that works.
+
+General principles:
+- No speculative abstractions, dependencies, classes, or boilerplate. Prefer deletion over addition, boring code over clever code.
+- When two approaches are similarly small, choose the one with correct edge-case behavior.
+- This economy does **not** apply to validation at trust boundaries, errors preventing silent data corruption, security, or scientific correctness.
+- Non-trivial new logic must leave one small runnable check that fails if the behavior breaks.
+- Reuse existing functions/classes/patterns before writing new ones.
+- Strictly avoid one-line / single-use helpers — inline unless extraction clearly improves readability.
+- Prefer f-strings. Add a short docstring to new/modified functions. Start new executable modules with a module docstring.
+- Avoid adding new classes unless asked; if one seems necessary, ask first.
+
+## Testing
+- Think about the test before implementing. Add a focused test or assert-based self-check for new logic.
+- Tests are deterministic (RNG seeded); run a single test with `uv run pytest <path>::<name>`.
+
+## Git
+- Don't run interactive git, and don't commit or push unless asked.
+
+## Boundaries
+- Don't modify the user's environment or add dependencies without explicit approval.
+- Don't call functions from `gw/` into `core/` (keep `core/` domain-agnostic).
\ No newline at end of file

From b578d9dc891030d1da04988e26d2f86ded12e931 Mon Sep 17 00:00:00 2001
From: Heinrich Campe <49278231+hcampe@users.noreply.github.com>
Date: Tue, 30 Jun 2026 16:32:16 +0200
Subject: [PATCH 04/15] more work

---
 dingo/core/dataset.py                         |   9 +-
 dingo/gw/dataset/generate_dataset.py          | 111 ++++++------------
 dingo/gw/dataset/waveform_dataset.py          |   7 +-
 .../waveform_generator/waveform_generator.py  |   7 +-
 4 files changed, 52 insertions(+), 82 deletions(-)

diff --git a/dingo/core/dataset.py b/dingo/core/dataset.py
index 49f9718b8..e15a42fac 100644
--- a/dingo/core/dataset.py
+++ b/dingo/core/dataset.py
@@ -1,11 +1,14 @@
 from typing import List, Optional, Union
 import ast
+import logging
 import h5py
 import numpy as np
 import pandas as pd
 
 from dingo.core.utils.misc import get_version
 
+log = logging.getLogger(__name__)
+
 
 def recursive_hdf5_save(group, d):
     for k, v in d.items():
@@ -136,7 +139,7 @@ def __init__(
             self.from_dictionary(dictionary)
 
     def to_file(self, file_name: str, mode: str = "w"):
-        print("Saving dataset to " + str(file_name))
+        log.info(f"Saving dataset to {file_name}")
         save_dict = {
             k: v
             for k, v in vars(self).items()
@@ -150,9 +153,9 @@ def to_file(self, file_name: str, mode: str = "w"):
                 f.attrs["dataset_type"] = self.dataset_type
 
     def from_file(self, file_name: str):
-        print(f"Loading dataset from {str(file_name)}.")
+        log.info(f"Loading dataset from {file_name}.")
         if self._leave_on_disk_keys:
-            print(f"Omitting data keys {self._leave_on_disk_keys}.")
+            log.info(f"Omitting data keys {self._leave_on_disk_keys}.")
 
         with h5py.File(file_name, "r") as f:
             loaded_dict = recursive_hdf5_load(
diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index a52ca9865..0b4fda8a0 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -1,30 +1,30 @@
-import argparse
 import copy
-import textwrap
+import logging
 from multiprocessing import Pool
 from pathlib import Path
 from typing import Dict, Tuple
 from functools import partial
 
+import hydra
 import numpy as np
 import pandas as pd
-import yaml
 from bilby.gw.prior import BBHPriorDict
+from hydra.utils import instantiate
+from omegaconf import DictConfig, OmegaConf
 from threadpoolctl import threadpool_limits
 from torchvision.transforms import Compose
 
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
-from dingo.gw.domains import build_domain
-from dingo.gw.prior import build_prior_with_defaults
 from dingo.gw.SVD import ApplySVD, SVDBasis
 from dingo.gw.transforms import WhitenFixedASD
 from dingo.gw.waveform_generator import (
-    NewInterfaceWaveformGenerator,
     WaveformGenerator,
     generate_waveforms_parallel,
 )
 from dingo.core.utils.misc import call_func_strict_output_dim
 
+log = logging.getLogger(__name__)
+
 
 def generate_parameters_and_polarizations(
     waveform_generator: WaveformGenerator,
@@ -47,7 +47,7 @@ def generate_parameters_and_polarizations(
     pandas DataFrame of parameters
     dictionary of numpy arrays corresponding to waveform polarizations
     """
-    print("Generating dataset of size " + str(num_samples))
+    log.info("Generating dataset of size " + str(num_samples))
     parameters = pd.DataFrame(prior.sample(num_samples))
 
     if num_processes > 1:
@@ -67,12 +67,12 @@ def generate_parameters_and_polarizations(
         polarizations_ok = {k: v[idx_ok] for k, v in polarizations.items()}
         parameters_ok = parameters.iloc[idx_ok]
         failed_percent = 100 * len(idx_failed) / len(parameters)
-        print(
+        log.info(
             f"{len(idx_failed)} out of {len(parameters)} configuration ({failed_percent:.1f}%) failed to generate."
         )
         with pd.option_context("display.max_rows", None, "display.max_columns", None):
-            print(parameters.iloc[idx_failed])
-        print(
+            log.info(parameters.iloc[idx_failed])
+        log.info(
             f"Only returning the {len(idx_ok)} successfully generated configurations."
         )
         return parameters_ok, polarizations_ok
@@ -115,7 +115,7 @@ def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int):
     )
     test_parameters.reset_index(drop=True, inplace=True)
 
-    print("Building SVD basis.")
+    log.info("Building SVD basis.")
     basis = SVDBasis()
     basis.generate_basis(train_data, size)
 
@@ -139,39 +139,36 @@ def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int):
     return basis, n_train, n_test
 
 
-def generate_dataset(settings: Dict, num_processes: int) -> WaveformDataset:
+def _dataset_settings(cfg: DictConfig) -> Dict:
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    settings.pop("out_file", None)
+    settings.pop("num_processes", None)
+    return settings
+
+
+def generate_dataset(cfg: DictConfig) -> WaveformDataset:
     """
     Generate a waveform dataset.
 
     Parameters
     ----------
-    settings : dict
-        Dictionary of settings to configure the dataset
-    num_processes : int
+    cfg : DictConfig
+        Hydra configuration for dataset generation.
 
     Returns
     -------
     A WaveformDataset based on the settings.
     """
 
-    prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    domain = build_domain(settings["domain"])
-
-    new_interface_flag = settings["waveform_generator"].get("new_interface", False)
-    if new_interface_flag:
-        waveform_generator = NewInterfaceWaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
-    else:
-        waveform_generator = WaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
+    prior = instantiate(cfg.intrinsic_prior, _convert_="all")
+    domain = instantiate(cfg.domain)
+    waveform_generator = instantiate(cfg.waveform_generator, domain=domain)
+    num_processes = cfg.get("num_processes", 1)
 
+    settings = _dataset_settings(cfg)
     dataset_dict = {"settings": settings}
 
-    if "compression" in settings:
+    if settings.get("compression", None) is not None:
         compression_transforms = []
 
         if "whitening" in settings["compression"]:
@@ -246,63 +243,25 @@ def generate_dataset(settings: Dict, num_processes: int) -> WaveformDataset:
     dataset_dict["parameters"] = parameters
     dataset_dict["polarizations"] = polarizations
 
-    dataset_dict[settings["num_samples"]] = len(parameters)
     dataset = WaveformDataset(dictionary=dataset_dict)
     return dataset
 
 
-def parse_args():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        description=textwrap.dedent(
-            """\
-        Generate a waveform dataset based on a settings file.
-        """
-        ),
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        required=True,
-        help="YAML file containing database settings",
-    )
-    parser.add_argument(
-        "--num_processes",
-        type=int,
-        default=1,
-        help="Number of processes to use in pool for parallel waveform generation",
-    )
-    parser.add_argument(
-        "--out_file",
-        type=str,
-        default="waveform_dataset.hdf5",
-        help="Name of file for storing dataset.",
-    )
-    return parser.parse_args()
-
-
-def _generate_dataset_main(
-    settings_file: str, out_file: str, num_processes: int
-) -> None:
-    if not Path(settings_file).is_file():
-        raise FileNotFoundError(f"dataset generation, failed to find {settings_file}")
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs/gw/dataset",
+    config_name="generate_dataset",
+)
+def main(cfg: DictConfig) -> None:
+    out_file = cfg.out_file
     if not Path(out_file).parent.is_dir():
         raise FileNotFoundError(
             f"dataset generation: can not create {out_file}: "
             f"{Path(out_file).parent} does not exist"
         )
-    # Load settings
-    with open(settings_file, "r") as f:
-        settings = yaml.safe_load(f)
-
-    dataset = generate_dataset(settings, num_processes)
+    dataset = generate_dataset(cfg)
     dataset.to_file(str(out_file))
 
 
-def main() -> None:
-    args = parse_args()
-    _generate_dataset_main(args.settings_file, args.out_file, args.num_processes)
-
-
 if __name__ == "__main__":
     main()
diff --git a/dingo/gw/dataset/waveform_dataset.py b/dingo/gw/dataset/waveform_dataset.py
index af750f5be..43d298c5e 100644
--- a/dingo/gw/dataset/waveform_dataset.py
+++ b/dingo/gw/dataset/waveform_dataset.py
@@ -2,6 +2,7 @@
 import h5py
 import numpy as np
 import torch.utils.data
+from hydra.utils import instantiate
 from torchvision.transforms import Compose
 
 from dingo.core.dataset import DingoDataset, recursive_hdf5_load
@@ -100,7 +101,11 @@ def load_supplemental(
         svd_size_update : int
             If provided, reduces the SVD size when decompressing (for speed).
         """
-        self.domain = build_domain(self.settings["domain"])
+        domain_settings = self.settings["domain"]
+        if "_target_" in domain_settings:
+            self.domain = instantiate(domain_settings)
+        else:
+            self.domain = build_domain(domain_settings)
 
         # We always call update_domain() (even if domain_update is None) because we
         # want to be sure that the data are consistent with the saved settings. In
diff --git a/dingo/gw/waveform_generator/waveform_generator.py b/dingo/gw/waveform_generator/waveform_generator.py
index 43f6a6488..f673cf699 100644
--- a/dingo/gw/waveform_generator/waveform_generator.py
+++ b/dingo/gw/waveform_generator/waveform_generator.py
@@ -1,4 +1,5 @@
 from functools import partial
+import logging
 from multiprocessing import Pool
 from math import isclose
 
@@ -31,6 +32,8 @@
 )
 from dingo.gw.transforms.waveform_transforms import DecimateAll
 
+log = logging.getLogger(__name__)
+
 
 class WaveformGenerator:
     """Generate polarizations using LALSimulation routines in the specified domain for a
@@ -142,12 +145,12 @@ def spin_conversion_phase(self):
     @spin_conversion_phase.setter
     def spin_conversion_phase(self, value):
         if value is None:
-            print(
+            log.info(
                 "Setting spin_conversion_phase = None. Using phase parameter for "
                 "conversion to cartesian spins."
             )
         else:
-            print(
+            log.info(
                 f"Setting spin_conversion_phase = {value}. Using this value for the "
                 f"phase parameter for conversion to cartesian spins."
             )

From 345b2ccfd5cecdbe398115951571ea4de26135c8 Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Wed, 1 Jul 2026 09:16:35 +0200
Subject: [PATCH 05/15] WIP: Import logger and replace print statements with
 logging calls (issue #152)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 dingo/core/dataset.py                         |  7 +--
 dingo/core/posterior_models/base_model.py     | 19 +++----
 dingo/core/result.py                          | 40 +++++++-------
 dingo/core/samplers.py                        | 30 +++++------
 dingo/core/utils/condor_utils.py              | 10 ++--
 dingo/core/utils/pt_to_hdf5.py                |  4 +-
 dingo/core/utils/trainutils.py                | 45 +++++++++-------
 dingo/gw/SVD.py                               | 29 +++++++----
 dingo/gw/conversion/spin_conversion.py        |  4 +-
 dingo/gw/data/data_download.py                |  3 +-
 dingo/gw/data/data_preparation.py             |  7 +--
 .../gw/dataset/evaluate_multibanded_domain.py | 31 +++++------
 dingo/gw/dataset/generate_dataset.py          | 11 ++--
 dingo/gw/dataset/generate_dataset_dag.py      |  4 +-
 dingo/gw/dataset/utils.py                     |  7 +--
 dingo/gw/download_strain_data.py              |  7 +--
 dingo/gw/importance_sampling/diagnostics.py   |  3 +-
 .../importance_sampling/importance_weights.py | 13 ++---
 dingo/gw/inference/gw_samplers.py             |  7 +--
 dingo/gw/injection.py                         |  5 +-
 dingo/gw/likelihood.py                        | 13 ++---
 dingo/gw/noise/asd_dataset.py                 |  9 ++--
 dingo/gw/noise/generate_dataset.py            |  7 +--
 dingo/gw/noise/synthetic/asd_sampling.py      |  3 +-
 dingo/gw/noise/utils.py                       |  3 +-
 dingo/gw/result.py                            | 21 ++++----
 dingo/gw/training/train_builders.py           | 31 ++++++-----
 dingo/gw/training/train_pipeline.py           | 52 +++++++++----------
 dingo/gw/training/train_pipeline_condor.py    | 13 ++---
 dingo/gw/training/utils.py                    | 16 +++---
 dingo/gw/transforms/waveform_transforms.py    |  3 +-
 .../waveform_generator/waveform_generator.py  |  9 ++--
 dingo/pipe/dag_creator.py                     |  6 ++-
 dingo/pipe/dingo_result.py                    |  5 +-
 dingo/pipe/nodes/generation_node.py           |  4 +-
 dingo/pipe/nodes/pe_summary_node.py           |  6 ++-
 dingo/pipe/plot.py                            |  6 ++-
 dingo/pipe/pp_test.py                         |  4 +-
 38 files changed, 274 insertions(+), 223 deletions(-)

diff --git a/dingo/core/dataset.py b/dingo/core/dataset.py
index 49f9718b8..62893e765 100644
--- a/dingo/core/dataset.py
+++ b/dingo/core/dataset.py
@@ -4,6 +4,7 @@
 import numpy as np
 import pandas as pd
 
+from dingo.core.utils.logging_utils import logger
 from dingo.core.utils.misc import get_version
 
 
@@ -136,7 +137,7 @@ def __init__(
             self.from_dictionary(dictionary)
 
     def to_file(self, file_name: str, mode: str = "w"):
-        print("Saving dataset to " + str(file_name))
+        logger.info("Saving dataset to " + str(file_name))
         save_dict = {
             k: v
             for k, v in vars(self).items()
@@ -150,9 +151,9 @@ def to_file(self, file_name: str, mode: str = "w"):
                 f.attrs["dataset_type"] = self.dataset_type
 
     def from_file(self, file_name: str):
-        print(f"Loading dataset from {str(file_name)}.")
+        logger.info(f"Loading dataset from {str(file_name)}.")
         if self._leave_on_disk_keys:
-            print(f"Omitting data keys {self._leave_on_disk_keys}.")
+            logger.info(f"Omitting data keys {self._leave_on_disk_keys}.")
 
         with h5py.File(file_name, "r") as f:
             loaded_dict = recursive_hdf5_load(
diff --git a/dingo/core/posterior_models/base_model.py b/dingo/core/posterior_models/base_model.py
index 68508d199..5df191c3d 100755
--- a/dingo/core/posterior_models/base_model.py
+++ b/dingo/core/posterior_models/base_model.py
@@ -10,6 +10,7 @@
 
 import torch
 import dingo.core.utils as utils
+from dingo.core.utils.logging_utils import logger
 from torch.utils.data import Dataset
 import time
 import numpy as np
@@ -198,7 +199,7 @@ def network_to_device(self, device):
         #     raise NotImplementedError('This needs testing!')
         #     # dim = 0 [512, ...] -> [256, ...], [256, ...] on 2 GPUs
         #     self.network = torch.nn.DataParallel(self.network)
-        print(f"Putting posterior model to device {self.device}.")
+        logger.info(f"Putting posterior model to device {self.device}.")
         self.network.to(self.device)
 
     def initialize_optimizer_and_scheduler(self):
@@ -394,7 +395,7 @@ def train(
 
         if test_only:
             test_loss = test_epoch(self, test_loader)
-            print(f"test loss: {test_loss:.3f}")
+            logger.info(f"test loss: {test_loss:.3f}")
 
         else:
             while not runtime_limits.limits_exceeded(self.epoch):
@@ -403,24 +404,24 @@ def train(
                 # Training
                 lr = utils.get_lr(self.optimizer)
                 with threadpool_limits(limits=1, user_api="blas"):
-                    print(f"\nStart training epoch {self.epoch} with lr {lr}")
+                    logger.info(f"Start training epoch {self.epoch} with lr {lr}")
                     time_start = time.time()
                     train_loss = train_epoch(self, train_loader)
                     train_time = time.time() - time_start
 
-                    print(
+                    logger.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(train_time, 60)
                         )
                     )
 
                     # Testing
-                    print(f"Start testing epoch {self.epoch}")
+                    logger.info(f"Start testing epoch {self.epoch}")
                     time_start = time.time()
                     test_loss = test_epoch(self, test_loader)
                     test_time = time.time() - time_start
 
-                    print(
+                    logger.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(time.time() - time_start, 60)
                         )
@@ -449,7 +450,7 @@ def train(
                             }
                         )
                     except ImportError:
-                        print("wandb not installed. Skipping logging to wandb.")
+                        logger.warning("wandb not installed. Skipping logging to wandb.")
 
                 if early_stopping is not None:
                     # Whether to use train or test loss
@@ -464,9 +465,9 @@ def train(
                             join(train_dir, "best_model.pt"), save_training_info=False
                         )
                     if early_stopping.early_stop:
-                        print("Early stopping")
+                        logger.info("Early stopping")
                         break
-                print(f"Finished training epoch {self.epoch}.\n")
+                logger.info(f"Finished training epoch {self.epoch}.")
 
 
 def train_epoch(pm, dataloader):
diff --git a/dingo/core/result.py b/dingo/core/result.py
index 67017271a..026441de4 100644
--- a/dingo/core/result.py
+++ b/dingo/core/result.py
@@ -16,6 +16,7 @@
 from bilby.core.prior import Constraint, DeltaFunction, PriorDict
 
 from dingo.core.dataset import DingoDataset
+from dingo.core.utils.logging_utils import logger
 from dingo.core.density import train_unconditional_density_estimator
 from dingo.core.utils.misc import recursive_check_dicts_are_equal
 from dingo.core.utils.plotting import get_latex_labels, plot_corner_multi
@@ -167,23 +168,24 @@ def reset_event(self, event_dataset):
         ):
             # This is really just for notification. Actions are only taken if the
             # event metadata differ.
-            print("\nNew event data differ from existing.")
+            logger.warning("New event data differ from existing.")
         self.context = context
 
         if self.event_metadata is not None and self.event_metadata != event_metadata:
-            print("Changes")
-            print("=======")
+            logger.warning("Changes")
+            logger.warning("=======")
             old_minus_new = dict(freeze(self.event_metadata) - freeze(event_metadata))
-            print("Old event metadata:")
+            logger.warning("Old event metadata:")
             for k in sorted(old_minus_new):
-                print(f"  {k}:  {self.event_metadata[k]}")
+                logger.warning(f"  {k}:  {self.event_metadata[k]}")
 
             new_minus_old = dict(freeze(event_metadata) - freeze(self.event_metadata))
-            print("New event metadata:")
+            logger.warning("New event metadata:")
             if self.importance_sampling_metadata.get("updates") is None:
                 self.importance_sampling_metadata["updates"] = {}
             for k in sorted(new_minus_old):
-                print(f"  {k}:  {event_metadata[k]}")
+                logger.warning(f"  {k}:  {event_metadata[k]}")
+                self.importance_sampling_metadata["updates"][k] = event_metadata[k]
                 self.importance_sampling_metadata["updates"][k] = event_metadata[k]
 
             self._rebuild_domain(verbose=True)
@@ -312,12 +314,12 @@ def importance_sample(self, num_processes: int = 1, **likelihood_kwargs):
         valid_samples = np.isfinite(log_prior + delta_log_prob_target)
         theta = theta.iloc[valid_samples]
 
-        print(f"Calculating {len(theta)} likelihoods.")
+        logger.info(f"Calculating {len(theta)} likelihoods.")
         t0 = time.time()
         log_likelihood = self.likelihood.log_likelihood_multi(
             theta, num_processes=num_processes
         )
-        print(f"Done. This took {time.time() - t0:.2f} seconds.")
+        logger.info(f"Done. This took {time.time() - t0:.2f} seconds.")
 
         self.log_noise_evidence = self.likelihood.log_Zn
         self.samples["log_prior"] = log_prior
@@ -494,7 +496,7 @@ def rejection_sample(
         rng = np.random.default_rng(random_state)
         weights = self.samples["weights"].to_numpy(dtype=float)
 
-        print(
+        logger.info(
             f"Rejection sampling: {self.num_samples} samples, "
             f"ESS = {self.effective_sample_size:.0f} "
             f"(efficiency = {100 * self.sample_efficiency:.1f}%)"
@@ -526,7 +528,7 @@ def rejection_sample(
         unweighted = unweighted.drop(
             columns=["weights", "log_prob", "delta_log_prob_target"], errors="ignore"
         )
-        print(f"Produced {len(unweighted)} unweighted samples.")
+        logger.info(f"Produced {len(unweighted)} unweighted samples.")
         return unweighted
 
     def parameter_subset(self, parameters):
@@ -619,12 +621,12 @@ def print_summary(self):
         Display the number of samples, and (if importance sampling is complete) the log
         evidence and number of effective samples.
         """
-        print("Number of samples:", len(self.samples))
+        logger.info("Number of samples: %d", len(self.samples))
         if self.log_evidence is not None:
-            print(
+            logger.info(
                 f"Log(evidence): {self.log_evidence:.3f} +- {self.log_evidence_std:.3f}"
             )
-            print(
+            logger.info(
                 f"Effective samples {self.n_eff:.1f}: "
                 f"(Sample efficiency = {100 * self.sample_efficiency:.2f}%)"
             )
@@ -815,7 +817,7 @@ def plot_log_probs(self, filename="log_probs.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            print("Results not importance sampled. Cannot produce log_prob plot.")
+            logger.warning("Results not importance sampled. Cannot produce log_prob plot.")
 
     def plot_weights(self, filename="weights.png"):
         """Make a scatter plot of samples weights vs log proposal."""
@@ -844,7 +846,7 @@ def plot_weights(self, filename="weights.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            print("Results not importance sampled. Cannot plot weights.")
+            logger.warning("Results not importance sampled. Cannot plot weights.")
 
     def get_all_injection_credible_levels(
         self, keys: list[str] = None, weighted: bool = False
@@ -1020,7 +1022,7 @@ def make_pp_plot(
 
     pvalues = []
     latex_labels = get_latex_labels(results[0].prior)
-    print("Key: KS-test p-value")
+    logger.info("Key: KS-test p-value")
     for ii, key in enumerate(credible_levels):
         pp = np.array(
             [
@@ -1030,7 +1032,7 @@ def make_pp_plot(
         )
         pvalue = scipy.stats.kstest(credible_levels[key], "uniform").pvalue
         pvalues.append(pvalue)
-        print("{}: {}".format(key, pvalue))
+        logger.info("{}: {}".format(key, pvalue))
         label = "{} ({:2.3f})".format(latex_labels[key], pvalue)
         plt.plot(x_values, pp, lines[ii], label=label, **kwargs)
 
@@ -1040,7 +1042,7 @@ def make_pp_plot(
         pvalues=pvalues,
         names=list(credible_levels.keys()),
     )
-    print("Combined p-value: {}".format(pvals.combined_pvalue))
+    logger.info("Combined p-value: {}".format(pvals.combined_pvalue))
 
     if title:
         ax.set_title(
diff --git a/dingo/core/samplers.py b/dingo/core/samplers.py
index efc4fc1fa..ab12ddf17 100644
--- a/dingo/core/samplers.py
+++ b/dingo/core/samplers.py
@@ -11,6 +11,7 @@
 from torchvision.transforms import Compose
 
 from dingo.core.posterior_models import BasePosteriorModel
+from dingo.core.utils.logging_utils import logger
 from dingo.core.result import Result
 from dingo.core.result import DATA_KEYS as RESULT_DATA_KEYS
 from dingo.core.utils import torch_detach_to_cpu, IterationTracker
@@ -157,7 +158,7 @@ def _run_sampler(
             x = [x]
         else:
             if context is not None:
-                print("Unconditional model. Ignoring context.")
+                logger.warning("Unconditional model. Ignoring context.")
             x = []
 
         # For a normalizing flow, we get the log_prob for "free" when sampling,
@@ -205,7 +206,7 @@ def run_sampler(
         """
         self.samples = None
 
-        print(f"Running sampler to generate {num_samples} samples.")
+        logger.info(f"Running sampler to generate {num_samples} samples.")
         t0 = time.time()
         if not self.unconditional_model:
             if self.context is None:
@@ -229,7 +230,7 @@ def run_sampler(
         # correction for t_ref) and represent as DataFrame.
         self._post_process(samples)
         self.samples = pd.DataFrame(samples)
-        print(f"Done. This took {time.time() - t0:.1f} s.")
+        logger.info(f"Done. This took {time.time() - t0:.1f} s.")
         sys.stdout.flush()
 
     def log_prob(self, samples: pd.DataFrame | dict) -> np.ndarray:
@@ -430,8 +431,8 @@ def _run_sampler(
             init_samples = self.init_sampler._run_sampler(num_samples, context)
         else:
             if self.num_iterations == 1:
-                print(
-                    f"Warning: Removing initial outliers, but only carrying out "
+                logger.warning(
+                    f"Removing initial outliers, but only carrying out "
                     f"{self.num_iterations} GNPE iteration. This risks biasing "
                     f"results."
                 )
@@ -515,16 +516,15 @@ def _run_sampler(
                 p: x["extrinsic_parameters"][p] for p in self.gnpe_proxy_parameters
             }
 
-            print(
-                f"it {i}.\tmin pvalue: {self.iteration_tracker.pvalue_min:.3f}"
-                f"\tproxy mean: ",
-                *[f"{torch.mean(v).item():.5f}" for v in proxies.values()],
-                "\tproxy std:",
-                *[f"{torch.std(v).item():.5f}" for v in proxies.values()],
-                "\ttimes:",
-                time_sample_start - start_time,
-                time_sample_end - time_sample_start,
-                time.time() - time_sample_end,
+            proxy_means = " ".join(f"{torch.mean(v).item():.5f}" for v in proxies.values())
+            proxy_stds = " ".join(f"{torch.std(v).item():.5f}" for v in proxies.values())
+            logger.debug(
+                f"it {i}. min pvalue: {self.iteration_tracker.pvalue_min:.3f}"
+                f" proxy mean: {proxy_means}"
+                f" proxy std: {proxy_stds}"
+                f" times: {time_sample_start - start_time:.3f}"
+                f" {time_sample_end - time_sample_start:.3f}"
+                f" {time.time() - time_sample_end:.3f}"
             )
 
         #
diff --git a/dingo/core/utils/condor_utils.py b/dingo/core/utils/condor_utils.py
index 1b0e2bbeb..cc51de919 100644
--- a/dingo/core/utils/condor_utils.py
+++ b/dingo/core/utils/condor_utils.py
@@ -2,6 +2,8 @@
 from os.path import join
 import yaml
 
+from dingo.core.utils.logging_utils import logger
+
 
 def resubmit_condor_job(train_dir, train_settings, epoch):
     """
@@ -12,14 +14,14 @@ def resubmit_condor_job(train_dir, train_settings, epoch):
     :return:
     """
     if 'condor_settings' in train_settings:
-        print('Copying log files')
+        logger.info("Copying log files")
         copy_logfiles(train_dir, epoch=epoch)
 
         if epoch >= train_settings['train_settings']['runtime_limits'][
             'max_epochs_total']:
-            print('Training complete, job will not be resubmitted')
+            logger.info("Training complete, job will not be resubmitted")
         else:
-            print('Training incomplete, resubmitting job.')
+            logger.info("Training incomplete, resubmitting job.")
             create_submission_file_and_submit_job(train_dir)
 
 
@@ -76,7 +78,7 @@ def copy_logfiles(log_dir, epoch, name='info', suffixes=('.err','.log','.out')):
         try:
             copyfile(src, dest)
         except:
-            print('Could not copy ' + src)
+            logger.warning("Could not copy " + src)
 
 
 if __name__ == '__main__':
diff --git a/dingo/core/utils/pt_to_hdf5.py b/dingo/core/utils/pt_to_hdf5.py
index 06e6aa96c..c3a0f7e81 100644
--- a/dingo/core/utils/pt_to_hdf5.py
+++ b/dingo/core/utils/pt_to_hdf5.py
@@ -5,6 +5,8 @@
 import json
 import argparse
 
+from dingo.core.utils.logging_utils import logger
+
 
 def parse_args():
     parser = argparse.ArgumentParser(
@@ -32,7 +34,7 @@ def main():
     # This is required for use on CVMFS
     root, ext = os.path.splitext(args.out_file)
     out_file_name = f'{root}_v{args.model_version_number}{ext}'
-    print('Output will be written to', out_file_name)
+    logger.info(f'Output will be written to {out_file_name}')
 
     # Load data into CPU memory since we'll be saving it using CPU libraries
     d = torch.load(args.in_file, map_location=torch.device("cpu"))
diff --git a/dingo/core/utils/trainutils.py b/dingo/core/utils/trainutils.py
index bc7d12154..87de405d0 100644
--- a/dingo/core/utils/trainutils.py
+++ b/dingo/core/utils/trainutils.py
@@ -5,6 +5,8 @@
 import csv
 from typing import Literal
 
+from dingo.core.utils.logging_utils import logger
+
 
 class AvgTracker:
     def __init__(self):
@@ -85,7 +87,7 @@ def __call__(self, val_loss: float):
         elif score < self.best_score + self.delta:
             self.counter += 1
             if self.verbose:
-                print(f"EarlyStopping counter: {self.counter} out of {self.patience}")
+                logger.info(f"EarlyStopping counter: {self.counter} out of {self.patience}")
             if self.counter >= self.patience:
                 self.early_stop = True
             return False
@@ -124,23 +126,26 @@ def get_avg(self):
 
     def print_info(self, batch_idx):
         if batch_idx % self.print_freq == 0:
-            print(
-                "{} Epoch: {} [{}/{} ({:.0f}%)]".format(
+            td, td_avg = self.times["Dataloader"].x, self.times["Dataloader"].get_avg()
+            tn, tn_avg = self.times["Network"].x, self.times["Network"].get_avg()
+            logger.debug(
+                "{} Epoch: {} [{}/{} ({:.0f}%)]  "
+                "Loss: {:.3f} ({:.3f})  "
+                "Time Dataloader: {:.3f} ({:.3f})  "
+                "Time Network: {:.3f} ({:.3f})".format(
                     self.mode,
                     self.epoch,
                     min(batch_idx * self.batch_size, self.len_dataset),
                     self.len_dataset,
                     100.0 * batch_idx * self.batch_size / self.len_dataset,
-                ),
-                end="\t\t",
+                    self.loss,
+                    self.get_avg(),
+                    td,
+                    td_avg,
+                    tn,
+                    tn_avg,
+                )
             )
-            # print loss
-            print(f"Loss: {self.loss:.3f} ({self.get_avg():.3f})", end="\t\t")
-            # print computation times
-            td, td_avg = self.times["Dataloader"].x, self.times["Dataloader"].get_avg()
-            tn, tn_avg = self.times["Network"].x, self.times["Network"].get_avg()
-            print(f"Time Dataloader: {td:.3f} ({td_avg:.3f})", end="\t\t")
-            print(f"Time Network: {tn:.3f} ({tn_avg:.3f})")
 
 
 class RuntimeLimits:
@@ -195,8 +200,8 @@ def limits_exceeded(self, epoch: int = None):
         # check time limit for run
         if self.max_time_per_run is not None:
             if time.time() - self.time_start >= self.max_time_per_run:
-                print(
-                    f"Stop run: Time limit of {self.max_time_per_run} s " f"exceeded."
+                logger.info(
+                    f"Stop run: Time limit of {self.max_time_per_run} s exceeded."
                 )
                 return True
         # check epoch limit for run
@@ -204,14 +209,14 @@ def limits_exceeded(self, epoch: int = None):
             if epoch is None:
                 raise ValueError("epoch required")
             if epoch - self.epoch_start >= self.max_epochs_per_run:
-                print(
+                logger.info(
                     f"Stop run: Epoch limit of {self.max_epochs_per_run} per run reached."
                 )
                 return True
         # check total epoch limit
         if self.max_epochs_total is not None:
             if epoch >= self.max_epochs_total:
-                print(
+                logger.info(
                     f"Stop run: Total epoch limit of {self.max_epochs_total} reached."
                 )
                 return True
@@ -316,13 +321,13 @@ def save_model(pm, log_dir, model_prefix="model", checkpoint_epochs=None):
     """
     # save current model
     model_name = join(log_dir, f"{model_prefix}_latest.pt")
-    print(f"Saving model to {model_name}.", end=" ")
+    logger.info(f"Saving model to {model_name}.")
     pm.save_model(model_name, save_training_info=True)
-    print("Done.")
+    logger.info("Done.")
 
     # potentially copy model to a checkpoint
     if checkpoint_epochs is not None and pm.epoch % checkpoint_epochs == 0:
         model_name_cp = join(log_dir, f"{model_prefix}_{pm.epoch:03d}.pt")
-        print(f"Copy model to checkpoint {model_name_cp}.", end=" ")
+        logger.info(f"Copying model to checkpoint {model_name_cp}.")
         copyfile(model_name, model_name_cp)
-        print("Done.")
+        logger.info("Done.")
diff --git a/dingo/gw/SVD.py b/dingo/gw/SVD.py
index e77b7761d..561d60f9a 100644
--- a/dingo/gw/SVD.py
+++ b/dingo/gw/SVD.py
@@ -3,6 +3,7 @@
 import scipy
 from sklearn.utils.extmath import randomized_svd
 from dingo.core.dataset import DingoDataset
+from dingo.core.utils.logging_utils import logger
 
 class SVDBasis(DingoDataset):
 
@@ -155,15 +156,25 @@ def print_validation_summary(self):
                 if "mismatch" in col:
                     n = int(col.split(sep="=")[-1])
                     mismatches = self.mismatches[col]
-                    print(f"n = {n}")
-                    print("  Mean mismatch = {}".format(np.mean(mismatches)))
-                    print("  Standard deviation = {}".format(np.std(mismatches)))
-                    print("  Max mismatch = {}".format(np.max(mismatches)))
-                    print("  Median mismatch = {}".format(np.median(mismatches)))
-                    print("  Percentiles:")
-                    print("    99    -> {}".format(np.percentile(mismatches, 99)))
-                    print("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-                    print("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
+                    logger.info(
+                        f"n = {n}\n"
+                        "  Mean mismatch = {}\n"
+                        "  Standard deviation = {}\n"
+                        "  Max mismatch = {}\n"
+                        "  Median mismatch = {}\n"
+                        "  Percentiles:\n"
+                        "    99    -> {}\n"
+                        "    99.9  -> {}\n"
+                        "    99.99 -> {}".format(
+                            np.mean(mismatches),
+                            np.std(mismatches),
+                            np.max(mismatches),
+                            np.median(mismatches),
+                            np.percentile(mismatches, 99),
+                            np.percentile(mismatches, 99.9),
+                            np.percentile(mismatches, 99.99),
+                        )
+                    )
 
     def decompress(self, coefficients: np.ndarray):
         """
diff --git a/dingo/gw/conversion/spin_conversion.py b/dingo/gw/conversion/spin_conversion.py
index 9e4b481cf..f2a40b45b 100644
--- a/dingo/gw/conversion/spin_conversion.py
+++ b/dingo/gw/conversion/spin_conversion.py
@@ -8,6 +8,8 @@
 import lalsimulation as LS
 from threadpoolctl import threadpool_limits
 
+from dingo.core.utils.logging_utils import logger
+
 
 DINGO_PE_SPIN_PARAMETERS = (
     "theta_jn",
@@ -225,5 +227,5 @@ def _convert_phase(f_ref, sc_phase_old, sc_phase_new, sample):
 
         return sample_pe_new
     except RuntimeError:
-        print("Failed to convert spins. Saving sample unchanged.")
+        logger.warning("Failed to convert spins. Saving sample unchanged.")
         return sample
diff --git a/dingo/gw/data/data_download.py b/dingo/gw/data/data_download.py
index 31e6b1c92..c0e2f6ed1 100644
--- a/dingo/gw/data/data_download.py
+++ b/dingo/gw/data/data_download.py
@@ -3,6 +3,7 @@
 import pycbc.psd
 import math
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.gwutils import get_window
 
 
@@ -42,7 +43,7 @@ def download_psd(det, time_start, time_psd, window, f_s):
     # if strain for PSD data contains nan, shift segment for PSD
     if np.max(np.isnan(psd_strain)):
         dt = math.ceil(np.where(np.isnan(psd_strain))[0][-1] / f_s)
-        print(
+        logger.warning(
             f"Nan encountered in strain data for PSD estimation for detector {det}. "
             f"Shifting strain segment by {dt} seconds."
         )
diff --git a/dingo/gw/data/data_preparation.py b/dingo/gw/data/data_preparation.py
index 0b06b1602..4f355b8b4 100644
--- a/dingo/gw/data/data_preparation.py
+++ b/dingo/gw/data/data_preparation.py
@@ -4,6 +4,7 @@
 from gwpy.timeseries import TimeSeries
 
 from dingo.core.dataset import DingoDataset
+from dingo.core.utils.logging_utils import logger
 from dingo.core.utils.misc import recursive_check_dicts_are_equal
 from dingo.gw.data.data_download import download_raw_data
 from dingo.gw.gwutils import get_window
@@ -45,11 +46,11 @@ def load_raw_data(time_event, settings, event_dataset=None):
                         f"{dataset.settings}"
                     )
             data = vars(dataset)[event]
-            print(f"Data for event at {event} found in {event_dataset}.")
+            logger.info(f"Data for event at {event} found in {event_dataset}.")
             return data
 
     # if this did not work, download the data
-    print(f"Downloading data for event at {event}.")
+    logger.info(f"Downloading data for event at {event}.")
     data = download_raw_data(time_event, **settings)
 
     # optionally save this data to event_dataset
@@ -57,7 +58,7 @@ def load_raw_data(time_event, settings, event_dataset=None):
         dataset = DingoDataset(
             dictionary={event: data, "settings": settings}, data_keys=[event]
         )
-        print(f"Saving data for event at {event} to {event_dataset}.")
+        logger.info(f"Saving data for event at {event} to {event_dataset}.")
         dataset.to_file(event_dataset, mode="a")
 
     return data
diff --git a/dingo/gw/dataset/evaluate_multibanded_domain.py b/dingo/gw/dataset/evaluate_multibanded_domain.py
index b20a2fb43..484603c29 100644
--- a/dingo/gw/dataset/evaluate_multibanded_domain.py
+++ b/dingo/gw/dataset/evaluate_multibanded_domain.py
@@ -4,6 +4,7 @@
 import yaml
 from scipy.interpolate import interp1d
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.dataset import generate_parameters_and_polarizations
 from dingo.gw.domains import build_domain, MultibandedFrequencyDomain
 from dingo.gw.gwutils import get_mismatch
@@ -35,13 +36,13 @@ def _evaluate_multibanding_main(
     settings["intrinsic_prior"]["chirp_mass"] = prior["chirp_mass"].minimum
     # Rebuild prior with updated settings.
     prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    print("Prior")
+    logger.info("Prior")
     for k, v in prior.items():
-        print(f"{k}: {v}")
+        logger.info(f"{k}: {v}")
 
     domain = build_domain(settings["domain"])
-    print("\nDomain")
-    print(domain.domain_dict)
+    logger.info("\nDomain")
+    logger.info(domain.domain_dict)
 
     if not isinstance(domain, MultibandedFrequencyDomain):
         raise ValueError("Waveform dataset domain not a MultibandedFrequencyDomain.")
@@ -88,21 +89,21 @@ def _evaluate_multibanding_main(
                 asd_file="aLIGO_ZERO_DET_high_P_asd.txt",
             )
 
-    print("\nMismatches between UFD waveforms and MFD waveforms interpolated to MFD.")
-    print(
+    logger.info("\nMismatches between UFD waveforms and MFD waveforms interpolated to MFD.")
+    logger.info(
         "This is a conservative estimate of the MFD performance when training "
         "networks."
     )
     mismatches = np.concatenate([v for v in mismatches.values()])
-    print(f"num_samples = {num_samples}")
-    print("  Mean mismatch = {}".format(np.mean(mismatches)))
-    print("  Standard deviation = {}".format(np.std(mismatches)))
-    print("  Max mismatch = {}".format(np.max(mismatches)))
-    print("  Median mismatch = {}".format(np.median(mismatches)))
-    print("  Percentiles:")
-    print("    99    -> {}".format(np.percentile(mismatches, 99)))
-    print("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-    print("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
+    logger.info(f"num_samples = {num_samples}")
+    logger.info("  Mean mismatch = {}".format(np.mean(mismatches)))
+    logger.info("  Standard deviation = {}".format(np.std(mismatches)))
+    logger.info("  Max mismatch = {}".format(np.max(mismatches)))
+    logger.info("  Median mismatch = {}".format(np.median(mismatches)))
+    logger.info("  Percentiles:")
+    logger.info("    99    -> {}".format(np.percentile(mismatches, 99)))
+    logger.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
+    logger.info("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
 
 
 def parse_args():
diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index a52ca9865..873568c13 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -24,6 +24,7 @@
     generate_waveforms_parallel,
 )
 from dingo.core.utils.misc import call_func_strict_output_dim
+from dingo.core.utils.logging_utils import logger
 
 
 def generate_parameters_and_polarizations(
@@ -47,7 +48,7 @@ def generate_parameters_and_polarizations(
     pandas DataFrame of parameters
     dictionary of numpy arrays corresponding to waveform polarizations
     """
-    print("Generating dataset of size " + str(num_samples))
+    logger.info("Generating dataset of size " + str(num_samples))
     parameters = pd.DataFrame(prior.sample(num_samples))
 
     if num_processes > 1:
@@ -67,12 +68,12 @@ def generate_parameters_and_polarizations(
         polarizations_ok = {k: v[idx_ok] for k, v in polarizations.items()}
         parameters_ok = parameters.iloc[idx_ok]
         failed_percent = 100 * len(idx_failed) / len(parameters)
-        print(
+        logger.warning(
             f"{len(idx_failed)} out of {len(parameters)} configuration ({failed_percent:.1f}%) failed to generate."
         )
         with pd.option_context("display.max_rows", None, "display.max_columns", None):
-            print(parameters.iloc[idx_failed])
-        print(
+            logger.warning(parameters.iloc[idx_failed].to_string())
+        logger.warning(
             f"Only returning the {len(idx_ok)} successfully generated configurations."
         )
         return parameters_ok, polarizations_ok
@@ -115,7 +116,7 @@ def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int):
     )
     test_parameters.reset_index(drop=True, inplace=True)
 
-    print("Building SVD basis.")
+    logger.info("Building SVD basis.")
     basis = SVDBasis()
     basis.generate_basis(train_data, size)
 
diff --git a/dingo/gw/dataset/generate_dataset_dag.py b/dingo/gw/dataset/generate_dataset_dag.py
index 395fa8fa8..c24c74b32 100644
--- a/dingo/gw/dataset/generate_dataset_dag.py
+++ b/dingo/gw/dataset/generate_dataset_dag.py
@@ -5,6 +5,8 @@
 import yaml
 import copy
 
+from dingo.core.utils.logging_utils import logger
+
 # Fixed file names
 svd_fn = "svd.hdf5"
 settings_part_fn = "settings_part.yaml"
@@ -284,7 +286,7 @@ def main():
         pass
 
     dagman.build()
-    print(f"DAG submission file written to {args.submit}.")
+    logger.info(f"DAG submission file written to {args.submit}.")
 
 
 if __name__ == "__main__":
diff --git a/dingo/gw/dataset/utils.py b/dingo/gw/dataset/utils.py
index e8911f87c..aa5e84855 100644
--- a/dingo/gw/dataset/utils.py
+++ b/dingo/gw/dataset/utils.py
@@ -6,6 +6,7 @@
 import yaml
 from typing import List
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.SVD import SVDBasis
 from dingo.gw.dataset.generate_dataset import train_svd_basis
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
@@ -26,7 +27,7 @@ def merge_datasets(dataset_list: List[WaveformDataset]) -> WaveformDataset:
     WaveformDataset containing the merged data.
     """
 
-    print(f"Merging {len(dataset_list)} datasets into one.")
+    logger.info(f"Merging {len(dataset_list)} datasets into one.")
 
     # This ensures that all the keys are copied into the new dataset. The "extensive"
     # parts of the dataset (parameters, waveforms) will be overwritten by the combined
@@ -103,7 +104,7 @@ def merge_datasets_cli():
         merged_dataset.settings = settings
 
     merged_dataset.to_file(args.out_file)
-    print(
+    logger.info(
         f"Complete. New dataset consists of {merged_dataset.settings['num_samples']} "
         f"samples."
     )
@@ -151,7 +152,7 @@ def build_svd_cli():
     basis, n_train, n_test = train_svd_basis(dataset, args.size, n_train)
     # FIXME: This is not an ideal treatment. We should update the waveform generation
     #  to always provide the requested number of waveforms.
-    print(
+    logger.info(
         f"SVD basis trained based on {n_train} waveforms and validated on {n_test} "
         f"waveforms. Note that if this differs from number requested, it will not be "
         f"reflected in the settings file. This is likely due to EOB failure to "
diff --git a/dingo/gw/download_strain_data.py b/dingo/gw/download_strain_data.py
index 336954098..87aa55693 100644
--- a/dingo/gw/download_strain_data.py
+++ b/dingo/gw/download_strain_data.py
@@ -2,6 +2,7 @@
 import pycbc.psd
 from gwpy.timeseries import TimeSeries
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.domains import UniformFrequencyDomain
 from dingo.gw.gwutils import (
     get_window,
@@ -107,14 +108,14 @@ def download_strain_data_in_FD(det, time_event, time_segment, time_buffer, windo
         array with the frequency domain strain
     """
     # download strain data
-    print("Downloading strain data for event.", end=" ")
+    logger.info("Downloading strain data for event.")
     event_strain = TimeSeries.fetch_open_data(
         det,
         time_event + time_buffer - time_segment,
         time_event + time_buffer,
         cache=True,
     )
-    print("Done.")
+    logger.info("Done.")
 
     # transform to FD
     if type(window) == dict:
@@ -165,7 +166,7 @@ def download_event_data_in_FD(
     """
     data = {"waveform": {}, "asds": {}}
     for det in detectors:
-        print("Detector {:}:".format(det))
+        logger.info("Detector {:}:".format(det))
 
         data["waveform"][det] = download_strain_data_in_FD(
             det, time_event, time_segment, time_buffer, window
diff --git a/dingo/gw/importance_sampling/diagnostics.py b/dingo/gw/importance_sampling/diagnostics.py
index 39cc95bb5..c178502f8 100644
--- a/dingo/gw/importance_sampling/diagnostics.py
+++ b/dingo/gw/importance_sampling/diagnostics.py
@@ -3,6 +3,7 @@
 import math
 import pandas as pd
 import matplotlib.pyplot as plt
+from dingo.core.utils.logging_utils import logger
 from dingo.core.utils.plotting import plot_corner_multi
 from dingo.gw.result import Result
 
@@ -219,7 +220,7 @@ def plot_diagnostics(
     inds = np.where(weights > threshold)[0]
     theta_new = theta.loc[inds]
     weights_new = weights[inds]
-    print(
+    logger.info(
         f"Generating cornerplot with {len(theta_new)} out of {len(theta)} IS samples."
     )
 
diff --git a/dingo/gw/importance_sampling/importance_weights.py b/dingo/gw/importance_sampling/importance_weights.py
index 7f88fa67e..e6c60a482 100644
--- a/dingo/gw/importance_sampling/importance_weights.py
+++ b/dingo/gw/importance_sampling/importance_weights.py
@@ -9,6 +9,7 @@
 from os.path import dirname, join, isfile, exists
 import argparse
 
+from dingo.core.utils.logging_utils import logger
 from dingo.core.posterior_models import NormalizingFlowPosteriorModel
 from dingo.gw.result import Result
 from dingo.gw.inference.gw_samplers import GWSampler
@@ -100,20 +101,20 @@ def main():
             "path", join(args.outdir, f"nde-{event_name}.pt")
         )
         if isfile(nde_name):
-            print(f"Loading nde at {nde_name} for event {event_name}.")
+            logger.info(f"Loading nde at {nde_name} for event {event_name}.")
             nde = NormalizingFlowPosteriorModel(
                 model_filename=nde_name,
                 device=settings["nde"]["training"]["device"],
                 load_training_info=False,
             )
         else:
-            print(f"Training new nde for event {event_name}.")
+            logger.info(f"Training new nde for event {event_name}.")
             nde = result.train_unconditional_flow(
                 inference_parameters,
                 settings["nde"],
                 train_dir=args.outdir,
             )
-            print(f"Renaming trained nde model to {nde_name}.")
+            logger.info(f"Renaming trained nde model to {nde_name}.")
             rename(join(args.outdir, "model_latest.pt"), nde_name)
 
         # Step 1a: Sample from proposal.
@@ -162,10 +163,10 @@ def main():
     # print(np.std(log_evidences) / np.mean(log_evidences_std))
 
     if synthetic_phase:
-        print(f"Sampling synthetic phase.")
+        logger.info("Sampling synthetic phase.")
         result.sample_synthetic_phase(synthetic_phase_kwargs)
 
-    print(f"Importance sampling.")
+    logger.info("Importance sampling.")
     result.importance_sample(
         num_processes=settings.get("num_processes", 1),
         time_marginalization_kwargs=time_marginalization_kwargs,
@@ -179,7 +180,7 @@ def main():
     diagnostics_dir = join(args.outdir, "IS-diagnostics")
     if not exists(diagnostics_dir):
         makedirs(diagnostics_dir)
-    print("Plotting diagnostics.")
+    logger.info("Plotting diagnostics.")
     plot_diagnostics(
         result,
         diagnostics_dir,
diff --git a/dingo/gw/inference/gw_samplers.py b/dingo/gw/inference/gw_samplers.py
index 2a1b63d89..c7d8f5585 100644
--- a/dingo/gw/inference/gw_samplers.py
+++ b/dingo/gw/inference/gw_samplers.py
@@ -8,6 +8,7 @@
 from torchvision.transforms import Compose
 
 from dingo.core.samplers import Sampler, GNPESampler
+from dingo.core.utils.logging_utils import logger
 from dingo.core.transforms import GetItem, RenameKey
 from dingo.gw.domains import (
     MultibandedFrequencyDomain,
@@ -240,7 +241,7 @@ def _post_process(self, samples: Union[dict, pd.DataFrame], inverse: bool = Fals
             for k, p in prior.items():
                 if isinstance(p, DeltaFunction) and k not in samples:
                     v = p.peak
-                    print(f"Adding fixed parameter {k} = {v} from prior.")
+                    logger.info(f"Adding fixed parameter {k} = {v} from prior.")
                     samples[k] = p.peak * np.ones(num_samples)
         else:
             # Drop non-inference parameters from samples.
@@ -472,8 +473,8 @@ def _initialize_transforms(self):
                 self.gnpe_parameters += transform.input_parameter_names
                 for k, v in transform.kernel.items():
                     self.gnpe_kernel[k] = v
-        print("GNPE parameters: ", self.gnpe_parameters)
-        print("GNPE kernel: ", self.gnpe_kernel)
+        logger.info(f"GNPE parameters: {self.gnpe_parameters}")
+        logger.info(f"GNPE kernel: {self.gnpe_kernel}")
 
         self.transform_pre = Compose(transform_pre)
 
diff --git a/dingo/gw/injection.py b/dingo/gw/injection.py
index 87f7a0660..f1679fb47 100644
--- a/dingo/gw/injection.py
+++ b/dingo/gw/injection.py
@@ -2,6 +2,7 @@
 from bilby.gw.detector import InterferometerList
 from torchvision.transforms import Compose
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.domains import (
     UniformFrequencyDomain,
@@ -312,7 +313,7 @@ def asd(self, asd):
             if set(asd.asds.keys()) != set(ifo_names):
                 raise KeyError("ASDDataset ifos do not match signal.")
             if asd.domain.domain_dict != self.data_domain.domain_dict:
-                print("Updating ASDDataset domain to match data domain.")
+                logger.info("Updating ASDDataset domain to match data domain.")
                 domain_dict = self.data_domain.domain_dict
                 asd.update_domain(domain_dict)
         elif isinstance(asd, dict):
@@ -402,7 +403,7 @@ def injection(self, theta):
             raise ValueError("self.asd must be set in order to produce injections.")
 
         if self.whiten:
-            print("self.whiten was set to True. Resetting to False.")
+            logger.warning("self.whiten was set to True. Resetting to False.")
             self.whiten = False
 
         data = {}
diff --git a/dingo/gw/likelihood.py b/dingo/gw/likelihood.py
index 49b29941f..f48893ad2 100644
--- a/dingo/gw/likelihood.py
+++ b/dingo/gw/likelihood.py
@@ -22,6 +22,7 @@
 )
 from dingo.gw.domains import build_domain
 from dingo.gw.data.data_preparation import get_event_data_and_domain
+from dingo.core.utils.logging_utils import logger
 
 
 class StationaryGaussianGWLikelihood(GWSignal, Likelihood):
@@ -173,7 +174,7 @@ def __init__(
                 # use endpoint = False for grid, since phase = 0/2pi are equivalent
                 self.phase_grid = np.linspace(0, 2 * np.pi, n_grid, endpoint=False)
             else:
-                print("Using phase marginalization with (2,2) mode approximation.")
+                logger.info("Using phase marginalization with (2,2) mode approximation.")
 
         # Initialize calibration marginalization using the setter from GWSignal.
         self.calibration_marginalization_kwargs = calibration_marginalization_kwargs
@@ -870,15 +871,15 @@ def main():
         try:
             l = likelihood.log_prob(theta)
         except:
-            print(idx)
+            logger.warning(idx)
             l = float("nan")
         log_likelihoods.append(l)
     log_likelihoods = np.array(log_likelihoods)
     log_likelihoods = log_likelihoods[~np.isnan(log_likelihoods)]
-    print(f"mean: {np.mean(log_likelihoods)}")
-    print(f"std: {np.std(log_likelihoods)}")
-    print(f"max: {np.max(log_likelihoods)}")
-    print(f"min: {np.min(log_likelihoods)}")
+    logger.info(f"mean: {np.mean(log_likelihoods)}")
+    logger.info(f"std: {np.std(log_likelihoods)}")
+    logger.info(f"max: {np.max(log_likelihoods)}")
+    logger.info(f"min: {np.min(log_likelihoods)}")
 
 
 if __name__ == "__main__":
diff --git a/dingo/gw/noise/asd_dataset.py b/dingo/gw/noise/asd_dataset.py
index 10e9cf9ef..9e3d1fb02 100644
--- a/dingo/gw/noise/asd_dataset.py
+++ b/dingo/gw/noise/asd_dataset.py
@@ -4,6 +4,7 @@
 
 import numpy as np
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.domains import build_domain, UniformFrequencyDomain
 from dingo.gw.domains.base_frequency_domain import BaseFrequencyDomain
 from dingo.gw.gwutils import *
@@ -59,8 +60,8 @@ def __init__(
                     self.gps_times.pop(ifo)
 
         if "window_factor" in self.settings["domain_dict"]:
-            print(
-                "Warning: 'window_factor' is no longer used in ASDDataset. "
+            logger.warning(
+                "'window_factor' is no longer used in ASDDataset. "
                 "Removing from settings."
             )
             self.settings["domain_dict"].pop("window_factor")
@@ -135,14 +136,14 @@ def update_domain(self, domain_update):
             self.domain.domain_dict["type"] == "UniformFrequencyDomain"
             and domain_update["type"] == "MultibandedFrequencyDomain"
         ):
-            print("Updating ASD dataset to MultibandedFrequencyDomain.")
+            logger.info("Updating ASD dataset to MultibandedFrequencyDomain.")
             asd_dataset_decimated = {}
             mfd = build_domain(domain_update)
             ufd = mfd.base_domain
             if not check_domain_compatibility(self.asds, ufd):
                 # If the ASD length is not compatible with the new base UFD,
                 # first truncate it.
-                print(
+                logger.info(
                     f"  Truncating first to new base UniformFrequencyDomain: f_max "
                     f"{self.domain.f_max} Hz -> {ufd.f_max} Hz"
                 )
diff --git a/dingo/gw/noise/generate_dataset.py b/dingo/gw/noise/generate_dataset.py
index e8750e6e2..2a4c8b383 100644
--- a/dingo/gw/noise/generate_dataset.py
+++ b/dingo/gw/noise/generate_dataset.py
@@ -12,6 +12,7 @@
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.generate_dataset_dag import create_dag
 from dingo.gw.noise.utils import merge_datasets, get_time_segments
+from dingo.core.utils.logging_utils import logger
 
 
 def parse_args():
@@ -94,11 +95,11 @@ def generate_dataset():
             pass
 
         dagman.build()
-        print(f"DAG submission file written.")
+        logger.info("DAG submission file written.")
 
     else:
 
-        print("Downloading strain data and estimating PSDs...")
+        logger.info("Downloading strain data and estimating PSDs...")
         asd_filename_list = download_and_estimate_psds(
             args.data_dir, settings, time_segments, verbose=args.verbose
         )
@@ -106,7 +107,7 @@ def generate_dataset():
             det: [ASDDataset(asd_file) for asd_file in asd_file_list]
             for det, asd_file_list in asd_filename_list.items()
         }
-        print("Merging single dataset files into one...")
+        logger.info("Merging single dataset files into one...")
         dataset = merge_datasets(asd_dataset_list)
         filename = args.out_name
         if filename is None:
diff --git a/dingo/gw/noise/synthetic/asd_sampling.py b/dingo/gw/noise/synthetic/asd_sampling.py
index 707ca586d..e5bfef1a6 100644
--- a/dingo/gw/noise/synthetic/asd_sampling.py
+++ b/dingo/gw/noise/synthetic/asd_sampling.py
@@ -2,6 +2,7 @@
 
 import numpy as np
 from scipy import stats
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.noise.synthetic.asd_parameterization import fit_broadband_noise
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.synthetic.utils import (
@@ -53,7 +54,7 @@ def fit(self, weights=None):
                         weights=weights,
                     )
                 except np.linalg.LinAlgError:
-                    print(
+                    logger.warning(
                         "Warning: Singular Matrix encountered in spectral KDE. Adding small Gaussian noise..."
                     )
                     perturbed_features = spectral_features[:, i, :] + np.random.normal(
diff --git a/dingo/gw/noise/utils.py b/dingo/gw/noise/utils.py
index 43d1f82ce..2d9fa4578 100644
--- a/dingo/gw/noise/utils.py
+++ b/dingo/gw/noise/utils.py
@@ -12,6 +12,7 @@
 import yaml
 
 from gwpy.table import EventTable
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.noise.asd_dataset import ASDDataset
 
 """
@@ -156,7 +157,7 @@ def merge_datasets(asd_dataset_list):
     merged_dict = {"asds": {}, "gps_times": {}}
 
     for det, asd_list in asd_dataset_list.items():
-        print(f"Merging {len(asd_list)} datasets into one for detector {det}.")
+        logger.info(f"Merging {len(asd_list)} datasets into one for detector {det}.")
         merged_dict["asds"][det] = np.vstack(
             [asd_dataset.asds[det] for asd_dataset in asd_list]
         )
diff --git a/dingo/gw/result.py b/dingo/gw/result.py
index 51616032d..b0499ce28 100644
--- a/dingo/gw/result.py
+++ b/dingo/gw/result.py
@@ -14,6 +14,7 @@
 )
 from dingo.core.multiprocessing import apply_func_with_multiprocessing
 from dingo.core.result import Result as CoreResult
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.conversion import change_spin_conversion_phase
 from dingo.gw.domains import MultibandedFrequencyDomain
 from dingo.gw.domains import build_domain
@@ -199,18 +200,14 @@ def _rebuild_domain(self, verbose=False):
             )
 
             if verbose:
-                print("Rebuilding domain as follows:")
-                print(
-                    yaml.dump(
-                        domain_dict,
-                        default_flow_style=False,
-                        sort_keys=False,
-                    )
+                logger.info(
+                    "Rebuilding domain as follows:\n"
+                    + yaml.dump(domain_dict, default_flow_style=False, sort_keys=False)
                 )
             self.domain = build_domain(domain_dict)
         else:
             if verbose:
-                print("No domain updates found; domain not rebuilt.")
+                logger.info("No domain updates found; domain not rebuilt.")
 
     def _build_prior(self):
         """Build the prior based on model metadata. Called by __init__()."""
@@ -359,7 +356,7 @@ def _build_likelihood(
         if "updates" in self.importance_sampling_metadata:
             if "T" in self.importance_sampling_metadata["updates"]:
                 delta_f_new = 1 / self.importance_sampling_metadata["updates"]["T"]
-                print(
+                logger.info(
                     f'Updating waveform generation delta_f from {wfg_domain_dict["delta_f"]} to {delta_f_new}.'
                 )
                 wfg_domain_dict["delta_f"] = delta_f_new
@@ -452,7 +449,7 @@ def sample_calibration_parameters(self, calibration_sampling_kwargs: dict):
 
         # Sample calibration parameters and calculate log_prob
         num_samples = len(self.samples)
-        print(f"Sampling calibration parameters for {num_samples} samples.")
+        logger.info(f"Sampling calibration parameters for {num_samples} samples.")
 
         delta_log_prob = np.zeros(num_samples)
 
@@ -571,7 +568,7 @@ def sample_synthetic_phase(
             self.synthetic_phase_kwargs.get("num_processes", 1), num_valid_samples // 10
         )
 
-        print(f"Estimating synthetic phase for {num_valid_samples} samples.")
+        logger.info(f"Estimating synthetic phase for {num_valid_samples} samples.")
         t0 = time.time()
 
         if not inverse:
@@ -660,7 +657,7 @@ def sample_synthetic_phase(
             self.samples["log_prob"] = log_prob_array
             del self.samples["phase"]
 
-        print(f"Done. This took {time.time() - t0:.2f} s.")
+        logger.info(f"Done. This took {time.time() - t0:.2f} s.")
 
     def get_samples_bilby_phase(self, num_processes=1):
         """
diff --git a/dingo/gw/training/train_builders.py b/dingo/gw/training/train_builders.py
index ca62856bb..2642eff3d 100755
--- a/dingo/gw/training/train_builders.py
+++ b/dingo/gw/training/train_builders.py
@@ -6,6 +6,7 @@
 from threadpoolctl import threadpool_limits
 from bilby.gw.detector import InterferometerList
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.SVD import SVDBasis
 
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
@@ -82,9 +83,9 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
         List of sub-transforms to omit from the full composition.
     """
 
-    print(f"Setting train transforms.")
+    logger.info("Setting train transforms.")
     if omit_transforms is not None:
-        print("Omitting \n\t" + "\n\t".join([t.__name__ for t in omit_transforms]))
+        logger.info("Omitting: " + ", ".join([t.__name__ for t in omit_transforms]))
 
     # By passing the wfd domain when instantiating the noise dataset, this ensures the
     # domains will match. In particular, it truncates the ASD dataset beyond the new
@@ -142,9 +143,9 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
     # parameters.
     try:
         standardization_dict = data_settings["standardization"]
-        print("Using previously-calculated parameter standardizations.")
+        logger.info("Using previously-calculated parameter standardizations.")
     except KeyError:
-        print("Calculating new parameter standardizations.")
+        logger.info("Calculating new parameter standardizations.")
         standardization_dict = get_standardization_dict(
             extrinsic_prior_dict,
             wfd,
@@ -258,7 +259,7 @@ def build_svd_for_embedding_network(
         ],
     )
 
-    print("Generating waveforms for embedding network SVD initialization.")
+    logger.info("Generating waveforms for embedding network SVD initialization.")
     time_start = time.time()
     ifos = list(wfd[0]["waveform"].keys())
     waveform_len = len(wfd[0]["waveform"][ifos[0]])
@@ -296,25 +297,25 @@ def build_svd_for_embedding_network(
                 waveforms[ifo][lower : lower + n] = strains[:n]
             if lower + n == num_waveforms:
                 break
-    print(f"...done. This took {time.time() - time_start:.0f} s.")
+    logger.info(f"Done. This took {time.time() - time_start:.0f} s.")
 
     # Reset the standard sharing strategy.
     torch.multiprocessing.set_sharing_strategy(old_sharing_strategy)
 
-    print("Generating SVD basis for ifo:")
+    logger.info("Generating SVD basis for ifo:")
     time_start = time.time()
     basis_dict = {}
     for ifo in ifos:
         basis = SVDBasis()
         basis.generate_basis(waveforms[ifo][:num_training_samples], size)
         basis_dict[ifo] = basis
-        print(f"...{ifo} done.")
-    print(f"...this took {time.time() - time_start:.0f} s.")
+        logger.info(f"...{ifo} done.")
+    logger.info(f"Done. This took {time.time() - time_start:.0f} s.")
 
     if out_dir is not None:
-        print(f"Testing SVD basis matrices.")
+        logger.info("Testing SVD basis matrices.")
         for ifo, basis in basis_dict.items():
-            print(f"...{ifo}:")
+            logger.info(f"...{ifo}:")
             basis.compute_test_mismatches(
                 waveforms[ifo][num_training_samples:],
                 parameters=parameters.iloc[num_training_samples:].reset_index(
@@ -323,19 +324,17 @@ def build_svd_for_embedding_network(
                 verbose=True,
             )
             basis.to_file(os.path.join(out_dir, f"svd_{ifo}.hdf5"))
-    print("Done")
+    logger.info("Done.")
 
     # Return V matrices in standard order. Drop the elements below domain.min_idx,
     # since the neural network expects data truncated below these. The dropped elements
     # should be 0.
-    print(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
-    print("...V matrix shapes:")
+    logger.info(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
     V_rb_list = []
     for ifo in data_settings["detectors"]:
         V = basis_dict[ifo].V
         assert np.allclose(V[: wfd.domain.min_idx], 0)
         V = V[wfd.domain.min_idx :]
-        print("      " + str(V.shape))
+        logger.info(f"  V matrix shape for {ifo}: {V.shape}")
         V_rb_list.append(V)
-    print("\n")
     return V_rb_list
diff --git a/dingo/gw/training/train_pipeline.py b/dingo/gw/training/train_pipeline.py
index ac1420b0f..544f4b1ac 100644
--- a/dingo/gw/training/train_pipeline.py
+++ b/dingo/gw/training/train_pipeline.py
@@ -27,6 +27,7 @@
     build_train_and_test_loaders,
 )
 from dingo.core.utils.trainutils import EarlyStopping
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.dataset import WaveformDataset
 from dingo.core.posterior_models import BasePosteriorModel
 
@@ -61,20 +62,20 @@ def copy_files_to_local(
     if local_dir is not None:
         file_name = file_path.split("/")[-1]
         local_file_path = os.path.join(local_dir, file_name)
-        print(f"Copying file to {local_file_path}")
+        logger.info(f"Copying file to {local_file_path}")
         # Copy file
         start_time = time.time()
         shutil.copy(file_path, local_file_path)
         elapsed_time = time.time() - start_time
-        print("Done. This took {:2.0f}:{:2.0f} min.".format(*divmod(elapsed_time, 60)))
+        logger.info("Done. This took {:2.0f}:{:2.0f} min.".format(*divmod(elapsed_time, 60)))
     elif leave_keys_on_disk and is_condor:
-        print(
-            f"Warning: leave_waveforms_on_disk defaults to True, but local_cache_path is not specified. "
-            f"This means that the waveforms will be loaded during training from {local_file_path} ."
+        logger.warning(
+            f"leave_waveforms_on_disk defaults to True, but local_cache_path is not specified. "
+            f"This means that the waveforms will be loaded during training from {local_file_path}. "
             f"This can lead to unexpected long times for data loading during training due to network traffic. "
             f"To prevent this, specify 'local_cache_path = tmp' in the local settings or set "
             f"leave_waveforms_on_disk = False. However, the latter is not recommended for large datasets since "
-            f"it can lead to memory issues when loading the entire dataset into RAM. "
+            f"it can lead to memory issues when loading the entire dataset into RAM."
         )
 
     return local_file_path
@@ -122,7 +123,7 @@ def prepare_training_new(
 
     if train_settings["model"].get("embedding_kwargs", None):
         # First, build the SVD for seeding the embedding network.
-        print("\nBuilding SVD for initialization of embedding network.")
+        logger.info("Building SVD for initialization of embedding network.")
         initial_weights["V_rb_list"] = build_svd_for_embedding_network(
             wfd,
             train_settings["data"],
@@ -153,9 +154,8 @@ def prepare_training_new(
         "train_settings": train_settings,
     }
 
-    print("\nInitializing new posterior model.")
-    print("Complete settings:")
-    print(yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
+    logger.info("Initializing new posterior model.")
+    logger.info("Complete settings:\n" + yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
 
     pm = build_model_from_kwargs(
         settings=full_settings,
@@ -173,7 +173,7 @@ def prepare_training_new(
                 **local_settings["wandb"],
             )
         except ImportError:
-            print("WandB is enabled but not installed.")
+            logger.warning("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -226,7 +226,7 @@ def prepare_training_resume(
                 **local_settings["wandb"],
             )
         except ImportError:
-            print("WandB is enabled but not installed.")
+            logger.warning("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -278,7 +278,7 @@ def initialize_stage(
     if not resume:
         # New optimizer and scheduler. If we are resuming, these should have been
         # loaded from the checkpoint.
-        print("Initializing new optimizer and scheduler.")
+        logger.info("Initializing new optimizer and scheduler.")
         pm.optimizer_kwargs = stage["optimizer"]
         pm.scheduler_kwargs = stage["scheduler"]
         pm.initialize_optimizer_and_scheduler()
@@ -295,7 +295,7 @@ def initialize_stage(
             )
     n_grad = get_number_of_model_parameters(pm.network, (True,))
     n_nograd = get_number_of_model_parameters(pm.network, (False,))
-    print(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}\n")
+    logger.info(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}")
 
     return train_loader, test_loader
 
@@ -343,14 +343,12 @@ def train_stages(
         stage = stages[n]
 
         if pm.epoch == end_epochs[n] - stage["epochs"]:
-            print(f"\nBeginning training stage {n}. Settings:")
-            print(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            logger.info(f"Beginning training stage {n}. Settings:\n" + yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=False
             )
         else:
-            print(f"\nResuming training in stage {n}. Settings:")
-            print(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            logger.info(f"Resuming training in stage {n}. Settings:\n" + yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=True
             )
@@ -359,7 +357,7 @@ def train_stages(
             try:
                 early_stopping = EarlyStopping(**stage["early_stopping"])
             except Exception:
-                print(
+                logger.warning(
                     "Early stopping settings invalid. Please pass 'patience', 'delta', 'metric'"
                 )
                 raise
@@ -381,10 +379,10 @@ def train_stages(
 
         if pm.epoch == end_epochs[n]:
             save_file = os.path.join(train_dir, f"model_stage_{n}.pt")
-            print(f"Training stage complete. Saving to {save_file}.")
+            logger.info(f"Training stage complete. Saving to {save_file}.")
             pm.save_model(save_file, save_training_info=True)
         if runtime_limits.local_limits_exceeded(pm.epoch):
-            print("Local runtime limits reached. Ending program.")
+            logger.info("Local runtime limits reached. Ending program.")
             break
 
     if pm.epoch == end_epochs[-1]:
@@ -443,7 +441,7 @@ def train_local():
     os.makedirs(args.train_dir, exist_ok=True)
 
     if args.settings_file is not None:
-        print("Beginning new training run.")
+        logger.info("Beginning new training run.")
         with open(args.settings_file, "r") as fp:
             train_settings = yaml.safe_load(fp)
 
@@ -462,7 +460,7 @@ def train_local():
 
                     local_settings["wandb"]["id"] = wandb.util.generate_id()
                 except ImportError:
-                    print("wandb not installed, cannot generate run id.")
+                    logger.warning("wandb not installed, cannot generate run id.")
             yaml.dump(local_settings, f, default_flow_style=False, sort_keys=False)
 
         pm, wfd = prepare_training_new(train_settings, args.train_dir, local_settings)
@@ -470,7 +468,7 @@ def train_local():
     else:
         if not os.path.isfile(args.checkpoint):
             raise FileNotFoundError(f"Checkpoint not found: {args.checkpoint}")
-        print("Resuming training run.")
+        logger.info("Resuming training run.")
         with open(os.path.join(args.train_dir, "local_settings.yaml"), "r") as f:
             local_settings = yaml.safe_load(f)
         pm, wfd = prepare_training_resume(
@@ -482,11 +480,11 @@ def train_local():
 
     if complete:
         if args.exit_command:
-            print(
+            logger.info(
                 f"All training stages complete. Executing exit command: {args.exit_command}."
             )
             os.system(args.exit_command)
         else:
-            print("All training stages complete.")
+            logger.info("All training stages complete.")
     else:
-        print("Program terminated due to runtime limit.")
+        logger.info("Program terminated due to runtime limit.")
diff --git a/dingo/gw/training/train_pipeline_condor.py b/dingo/gw/training/train_pipeline_condor.py
index 418909d3e..95f118add 100644
--- a/dingo/gw/training/train_pipeline_condor.py
+++ b/dingo/gw/training/train_pipeline_condor.py
@@ -4,6 +4,7 @@
 import yaml
 import argparse
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.training import (
     prepare_training_new,
     prepare_training_resume,
@@ -61,7 +62,7 @@ def copy_logfiles(log_dir, epoch, name="info", suffixes=(".err", ".log", ".out")
         try:
             copyfile(src, dest)
         except:
-            print("Could not copy " + src)
+            logger.warning("Could not copy " + src)
 
 
 def train_condor():
@@ -90,7 +91,7 @@ def train_condor():
                 raise FileNotFoundError(
                     f"Checkpoint not found: {join(args.train_dir, args.checkpoint)}"
                 )
-            print("Beginning new training run.")
+            logger.info("Beginning new training run.")
             with open(join(args.train_dir, "train_settings.yaml"), "r") as f:
                 train_settings = yaml.safe_load(f)
 
@@ -109,7 +110,7 @@ def train_condor():
 
                         local_settings["wandb"]["id"] = wandb.util.generate_id()
                     except ImportError:
-                        print("wandb not installed, cannot generate run id.")
+                        logger.warning("wandb not installed, cannot generate run id.")
                 yaml.dump(local_settings, f, default_flow_style=False, sort_keys=False)
 
             pm, wfd = prepare_training_new(
@@ -117,7 +118,7 @@ def train_condor():
             )
 
         else:
-            print("Resuming training run.")
+            logger.info("Resuming training run.")
             with open(os.path.join(args.train_dir, "local_settings.yaml"), "r") as f:
                 local_settings = yaml.safe_load(f)
             pm, wfd = prepare_training_resume(
@@ -128,7 +129,7 @@ def train_condor():
 
         complete = train_stages(pm, wfd, args.train_dir, local_settings)
 
-        print("Copying log files")
+        logger.info("Copying log files")
         copy_logfiles(args.train_dir, epoch=pm.epoch)
 
         #
@@ -136,7 +137,7 @@ def train_condor():
         #
 
         if complete:
-            print(
+            logger.info(
                 f"Training complete, job will not be resubmitted. Executing exit command: {args.exit_command}."
             )
             if args.exit_command:
diff --git a/dingo/gw/training/utils.py b/dingo/gw/training/utils.py
index 4cd0732d8..2aeedc964 100644
--- a/dingo/gw/training/utils.py
+++ b/dingo/gw/training/utils.py
@@ -5,6 +5,7 @@
 import yaml
 
 from dingo.core.utils.backward_compatibility import torch_load_with_fallback
+from dingo.core.utils.logging_utils import logger
 
 
 def append_stage():
@@ -25,7 +26,7 @@ def append_stage():
         if k.startswith("stage_")
     ]
     num_stages = len(stages)
-    print(f"Checkpoint training plan consists of {num_stages} stages.")
+    logger.info(f"Checkpoint training plan consists of {num_stages} stages.")
 
     with open(args.stage_settings_file, "r") as f:
         new_stage = yaml.safe_load(f)
@@ -39,21 +40,20 @@ def append_stage():
         current_epoch = d["epoch"]
         stage_epoch = np.sum([s["epochs"] for s in stages[: args.replace]])
         if current_epoch > stage_epoch:
-            print(
-                f"WARNING: Modification to training plan changes a training stage "
+            logger.warning(
+                f"Modification to training plan changes a training stage "
                 f"that has already started. Current model epoch is {current_epoch}. "
                 f"Proceed at your own risk!"
             )
-        print(f"Replacing planned stage {args.replace} with new stage.")
+        logger.info(f"Replacing planned stage {args.replace} with new stage.")
         new_stage_number = args.replace
     else:
-        print(f"Appending new stage to training plan.")
+        logger.info("Appending new stage to training plan.")
         new_stage_number = num_stages
 
     d["metadata"]["train_settings"]["training"][f"stage_{new_stage_number}"] = new_stage
-    print("Summary of new training plan:")
-    print(
-        yaml.dump(
+    logger.info(
+        "Summary of new training plan:\n" + yaml.dump(
             d["metadata"]["train_settings"]["training"],
             default_flow_style=False,
             sort_keys=False,
diff --git a/dingo/gw/transforms/waveform_transforms.py b/dingo/gw/transforms/waveform_transforms.py
index f6d022435..39f884836 100644
--- a/dingo/gw/transforms/waveform_transforms.py
+++ b/dingo/gw/transforms/waveform_transforms.py
@@ -1,6 +1,7 @@
 from typing import Optional
 import numpy as np
 
+from dingo.core.utils.logging_utils import logger
 from dingo.gw.domains import MultibandedFrequencyDomain, UniformFrequencyDomain
 
 
@@ -443,7 +444,7 @@ def __init__(
         self.maximum_frequency = maximum_frequency
 
         if print_output:
-            print(
+            logger.info(
                 f"Transform MaskDataForFrequencyRangeUpdate activated:"
                 f"  Settings: \n"
                 f"    - Minimum_frequency update: {self.minimum_frequency}\n"
diff --git a/dingo/gw/waveform_generator/waveform_generator.py b/dingo/gw/waveform_generator/waveform_generator.py
index 43f6a6488..a7a341ee4 100644
--- a/dingo/gw/waveform_generator/waveform_generator.py
+++ b/dingo/gw/waveform_generator/waveform_generator.py
@@ -30,6 +30,7 @@
     TimeDomain,
 )
 from dingo.gw.transforms.waveform_transforms import DecimateAll
+from dingo.core.utils.logging_utils import logger
 
 
 class WaveformGenerator:
@@ -142,12 +143,12 @@ def spin_conversion_phase(self):
     @spin_conversion_phase.setter
     def spin_conversion_phase(self, value):
         if value is None:
-            print(
+            logger.info(
                 "Setting spin_conversion_phase = None. Using phase parameter for "
                 "conversion to cartesian spins."
             )
         else:
-            print(
+            logger.info(
                 f"Setting spin_conversion_phase = {value}. Using this value for the "
                 f"phase parameter for conversion to cartesian spins."
             )
@@ -583,7 +584,7 @@ def generate_FD_waveform(
         # numbers if multibanding is used. If that happens, turn off multibanding to
         # fix this.
         if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-            print(
+            logger.warning(
                 f"Generation with parameters {parameters_lal} likely numerically "
                 f"unstable due to multibanding, turn off multibanding."
             )
@@ -606,7 +607,7 @@ def generate_FD_waveform(
                 *parameters_lal[lal_dict_idx + 1 :],
             )
             if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-                print(
+                logger.warning(
                     f"Warning: turning off multibanding for parameters {parameters_lal}"
                     f" likely numerically might not have fixed it, check manually."
                 )
diff --git a/dingo/pipe/dag_creator.py b/dingo/pipe/dag_creator.py
index 077864265..1223a553b 100644
--- a/dingo/pipe/dag_creator.py
+++ b/dingo/pipe/dag_creator.py
@@ -5,7 +5,9 @@
 import copy
 
 from bilby_pipe.job_creation.dag import Dag
-from bilby_pipe.utils import BilbyPipeError, logger
+import logging
+
+from bilby_pipe.utils import BilbyPipeError
 
 from dingo.pipe.utils import _strip_unwanted_submission_keys
 from dingo.pipe.nodes.generation_node import GenerationNode
@@ -15,7 +17,7 @@
 from .nodes.plot_node import PlotNode, PlotPPNode
 from .nodes.sampling_node import SamplingNode
 
-logger.name = "dingo_pipe"
+logger = logging.getLogger("dingo.pipe")
 
 
 def get_trigger_time_list(inputs):
diff --git a/dingo/pipe/dingo_result.py b/dingo/pipe/dingo_result.py
index 4b0096e13..b06c1dfdd 100644
--- a/dingo/pipe/dingo_result.py
+++ b/dingo/pipe/dingo_result.py
@@ -1,8 +1,11 @@
 import argparse
+import logging
 from pathlib import Path
 
 from dingo.gw.result import Result
 
+logger = logging.getLogger("dingo.pipe")
+
 
 def main():
     parser = argparse.ArgumentParser()
@@ -15,7 +18,7 @@ def main():
     args = parser.parse_args()
 
     if args.merge:
-        print(f"Merging {len(args.result)} parts into complete Result.")
+        logger.info(f"Merging {len(args.result)} parts into complete Result.")
         sub_results = []
         for file_name in args.result:
             sub_results.append(Result(file_name=file_name))
diff --git a/dingo/pipe/nodes/generation_node.py b/dingo/pipe/nodes/generation_node.py
index edbd4d9b2..6f7cfcc57 100644
--- a/dingo/pipe/nodes/generation_node.py
+++ b/dingo/pipe/nodes/generation_node.py
@@ -1,3 +1,4 @@
+import logging
 import os
 
 from bilby_pipe.job_creation.nodes import GenerationNode as BilbyGenerationNode
@@ -5,6 +6,8 @@
 
 from dingo.pipe.utils import _strip_unwanted_submission_keys
 
+logger = logging.getLogger("dingo.pipe")
+
 
 class GenerationNode(BilbyGenerationNode):
 
@@ -101,7 +104,6 @@ def __init__(self, inputs, trigger_time, idx, dag, parent=None, importance_sampl
             for fname in self.extract_paths_from_dict(frame_files)
             if fname.startswith(self.inputs.data_find_urltype)
         ]:
-            from bilby_pipe.utils import logger
             logger.warning(
                 "The following frame files were identified by gwdatafind for this analysis. "
                 "These frames may not be found by the data generation stage as file "
diff --git a/dingo/pipe/nodes/pe_summary_node.py b/dingo/pipe/nodes/pe_summary_node.py
index 6762ece07..876e7a0d1 100644
--- a/dingo/pipe/nodes/pe_summary_node.py
+++ b/dingo/pipe/nodes/pe_summary_node.py
@@ -1,8 +1,10 @@
 import os 
 from bilby_pipe.job_creation.nodes import PESummaryNode as BilbyPESummaryNode
-from bilby_pipe.utils import BilbyPipeError, logger
+import logging
 
-logger.name = "dingo_pipe"
+from bilby_pipe.utils import BilbyPipeError
+
+logger = logging.getLogger("dingo.pipe")
 
 
 class PESummaryNode(BilbyPESummaryNode):
diff --git a/dingo/pipe/plot.py b/dingo/pipe/plot.py
index 7dc8af756..dbd69de0c 100644
--- a/dingo/pipe/plot.py
+++ b/dingo/pipe/plot.py
@@ -1,9 +1,11 @@
 from bilby_pipe.bilbyargparser import BilbyArgParser
-from bilby_pipe.utils import get_command_line_arguments, logger, parse_args
+from bilby_pipe.utils import get_command_line_arguments, parse_args
+
+import logging
 
 from dingo.gw.result import Result
 
-logger.name = "dingo_pipe"
+logger = logging.getLogger("dingo.pipe")
 
 
 def create_parser():
diff --git a/dingo/pipe/pp_test.py b/dingo/pipe/pp_test.py
index 0716b24e6..bbc85502d 100644
--- a/dingo/pipe/pp_test.py
+++ b/dingo/pipe/pp_test.py
@@ -1,12 +1,12 @@
 import argparse
 from pathlib import Path
 
-from bilby_pipe.utils import logger
+import logging
 
 from dingo.core.result import make_pp_plot
 from dingo.gw.result import Result
 
-logger.name = "dingo_pipe"
+logger = logging.getLogger("dingo.pipe")
 
 
 def create_parser():

From f78608fda3d11b33433dd2dfb87b91c7d1d9ab5d Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Wed, 1 Jul 2026 09:16:50 +0200
Subject: [PATCH 06/15] WIP: Integrate bilby and dingo loggers to share log
 file (issue #152)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 dingo/core/utils/logging_utils.py | 70 ++++++++++++++++++-------------
 1 file changed, 41 insertions(+), 29 deletions(-)

diff --git a/dingo/core/utils/logging_utils.py b/dingo/core/utils/logging_utils.py
index e626fce0b..0de9d14e8 100644
--- a/dingo/core/utils/logging_utils.py
+++ b/dingo/core/utils/logging_utils.py
@@ -2,12 +2,6 @@
 import os
 import sys
 
-"""
-Utility functions
-"""
-
-# This is not currently used. We should improve our logging throughout.
-
 
 def check_directory_exists_and_if_not_mkdir(directory, logger):
     """Checks if the given directory exists and creates it if it does not exist
@@ -16,8 +10,6 @@ def check_directory_exists_and_if_not_mkdir(directory, logger):
     ----------
     directory: str
         Name of the directory
-
-    Borrowed from bilby-pipe: https://git.ligo.org/lscsoft/bilby_pipe/
     """
     if not os.path.exists(directory):
         os.makedirs(directory)
@@ -27,7 +19,11 @@ def check_directory_exists_and_if_not_mkdir(directory, logger):
 
 
 def setup_logger(outdir=None, label=None, log_level="INFO"):
-    """Setup logging output: call at the start of the script to use
+    """Setup logging output: call at the start of the script to use.
+
+    Sets up the dingo logger and, if bilby is available, also calls bilby's
+    setup_logger so that both loggers share the same file handler and their
+    entries appear together in the same log file.
 
     Parameters
     ----------
@@ -37,10 +33,7 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
         ['debug', 'info', 'warning']
         Either a string from the list above, or an integer as specified
         in https://docs.python.org/2/library/logging.html#logging-levels
-
-    Borrowed from bilby-pipe: https://git.ligo.org/lscsoft/bilby_pipe/
     """
-
     if "-v" in sys.argv or "--verbose" in sys.argv:
         log_level = "DEBUG"
 
@@ -52,12 +45,25 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
     else:
         level = int(log_level)
 
+    # Set up bilby's logger first so we can share its file handler.
+    bilby_logger = None
+    try:
+        from bilby.core.utils.log import setup_logger as bilby_setup_logger
+
+        bilby_setup_logger(outdir=outdir or ".", label=label, log_level=log_level)
+        bilby_logger = logging.getLogger("bilby")
+    except ImportError:
+        pass
+
     logger = logging.getLogger("dingo")
     logger.propagate = False
     logger.setLevel(level)
 
-    streams = [isinstance(h, logging.StreamHandler) for h in logger.handlers]
-    if len(streams) == 0 or not all(streams):
+    # Add a stream handler if not already present.
+    if not any(
+        isinstance(h, logging.StreamHandler) and not isinstance(h, logging.FileHandler)
+        for h in logger.handlers
+    ):
         stream_handler = logging.StreamHandler()
         stream_handler.setFormatter(
             logging.Formatter(
@@ -67,22 +73,28 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
         stream_handler.setLevel(level)
         logger.addHandler(stream_handler)
 
-    if any([isinstance(h, logging.FileHandler) for h in logger.handlers]) is False:
-        if label:
-            if outdir:
-                check_directory_exists_and_if_not_mkdir(outdir, logger)
-            else:
-                outdir = "."
-            log_file = f"{outdir}/{label}.log"
-            file_handler = logging.FileHandler(log_file)
-            file_handler.setFormatter(
-                logging.Formatter(
-                    "%(asctime)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
-                )
+    # Share bilby's file handler so both loggers write to the same log file.
+    if bilby_logger is not None:
+        for handler in bilby_logger.handlers:
+            if isinstance(handler, logging.FileHandler):
+                if not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
+                    logger.addHandler(handler)
+
+    # Fall back to a standalone file handler if bilby is unavailable.
+    if label and not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
+        if outdir:
+            check_directory_exists_and_if_not_mkdir(outdir, logger)
+        else:
+            outdir = "."
+        log_file = f"{outdir}/{label}.log"
+        file_handler = logging.FileHandler(log_file)
+        file_handler.setFormatter(
+            logging.Formatter(
+                "%(asctime)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
             )
-
-            file_handler.setLevel(level)
-            logger.addHandler(file_handler)
+        )
+        file_handler.setLevel(level)
+        logger.addHandler(file_handler)
 
     for handler in logger.handlers:
         handler.setLevel(level)

From b02968ebd81b28d3aba95ed3f4ba1e37a8864b4b Mon Sep 17 00:00:00 2001
From: Heinrich Campe <49278231+hcampe@users.noreply.github.com>
Date: Wed, 1 Jul 2026 11:38:17 +0200
Subject: [PATCH 07/15] generate dataset w hydra

---
 .gitignore                             |   3 +
 HYDRA_TODO.md                          | 426 ++++++++++++++++++++++
 configs/examples/generate_dataset.yaml |  42 +++
 dingo/gw/dataset/generate_dataset.py   |  38 +-
 pyproject.toml                         |   1 +
 uv.lock                                | 483 +++++++++++++------------
 6 files changed, 757 insertions(+), 236 deletions(-)
 create mode 100644 HYDRA_TODO.md
 create mode 100644 configs/examples/generate_dataset.yaml

diff --git a/.gitignore b/.gitignore
index a6c7293c0..5329a843c 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,6 +1,9 @@
 # codex stuff
 AGENTS*.md
 
+# run dir
+my_runs/
+
 
 # Byte-compiled / optimized / DLL files
 __pycache__/
diff --git a/HYDRA_TODO.md b/HYDRA_TODO.md
new file mode 100644
index 000000000..ecc322c7a
--- /dev/null
+++ b/HYDRA_TODO.md
@@ -0,0 +1,426 @@
+# Hydra Migration TODO
+
+This document tracks a staged migration of Dingo from ad hoc YAML files,
+`argparse` / `configargparse`, and hand-written config builders toward Hydra.
+
+## Migration Philosophy
+
+Do this in stages rather than trying to refactor the whole configuration model in
+one pass.
+
+1. **Stage 1: Hydra wrappers with old internal dictionaries**
+   - Replace public CLI parsers with `@hydra.main(...)` entrypoints.
+   - Replace old example/settings YAML files with Hydra configs that are mostly
+     one-to-one copies of the old settings.
+   - Keep these Stage 1 configs together in one examples-like folder under
+     `configs/`, mirroring the current `examples/` organization as much as
+     possible.
+   - Convert `DictConfig` to plain dictionaries internally with
+     `OmegaConf.to_container(cfg, resolve=True)`.
+   - Let the existing implementation consume those dictionaries as before.
+   - Preserve settings that are saved into datasets, models, and results.
+   - Replace command-line `print(...)` status output with module loggers:
+     `log = logging.getLogger(__name__)` and `log.info(...)`.
+   - Remove the old argparse/configargparse command-line arguments for migrated
+     commands, as was done for `dingo_generate_dataset`.
+
+2. **Stage 2: Organize configs with Hydra groups and defaults**
+   - Split the Stage 1 examples-like config folder into reusable groups.
+   - Deduplicate examples and common settings.
+   - Keep code mostly dictionary-based.
+
+3. **Stage 3: Refactor internals to use Hydra-native construction**
+   - Replace builder functions and string switches with `_target_` and
+     `hydra.utils.instantiate(...)`.
+   - Make transform/model/domain/prior construction more composable.
+   - Move defaults that are currently hard-coded in Python into default Hydra
+     configs/config groups.
+   - Make configs complete: options that are currently omitted from example
+     YAMLs because code supplies implicit defaults should become explicit
+     defaults in Hydra configs.
+   - Revisit saved metadata format and backward compatibility deliberately.
+
+## Stage 1 Definition Of Done
+
+For each migrated CLI:
+
+- [ ] Public entrypoint is wrapped by `@hydra.main(...)`.
+- [ ] Old `argparse`, `configargparse`, or `BilbyArgParser` parsing is removed
+      from that entrypoint.
+- [ ] Config lives in one examples-like Stage 1 folder under `configs/`.
+- [ ] Existing settings YAMLs/examples have equivalent Hydra configs.
+- [ ] `.ini` files are explicitly out of scope for this migration pass.
+- [ ] The Hydra config is converted to a plain dict before entering most
+      existing business logic.
+- [ ] Settings saved into HDF5/PT/result metadata remain loadable.
+- [ ] User-facing progress messages use `logging` instead of `print`.
+- [ ] A small smoke test exists or is documented.
+
+Recommended shared pattern:
+
+```python
+import logging
+
+import hydra
+from omegaconf import DictConfig, OmegaConf
+
+log = logging.getLogger(__name__)
+
+
+@hydra.main(version_base="1.3", config_path="...", config_name="...")
+def main(cfg: DictConfig) -> None:
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    ...
+```
+
+Recommended default Hydra run-directory policy:
+
+```yaml
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: <job_name>
+    chdir: true
+```
+
+Use `hydra.job.chdir: true` in Stage 1 so all relative outputs from a run land
+in the Hydra run directory together with `generate_dataset.log` and `.hydra/`
+metadata. The run root is controlled by the required `DINGO_RUNS` environment
+variable.
+
+If a user explicitly wants old current-working-directory behavior, use both:
+
+```bash
+hydra.job.chdir=false hydra.run.dir=.
+```
+
+Testing showed that `hydra.job.chdir=false` alone is not enough: application
+outputs land in the current working directory, but Hydra logs and `.hydra/`
+metadata still go to `hydra.run.dir`.
+
+Use `compression: null` rather than `compression: None`.
+
+Do not migrate `.ini` files for now. In particular, leave Asimov/plugin-style
+`.ini` configuration outside the Hydra migration TODO unless this is reopened
+explicitly later.
+
+## Entrypoint Inventory
+
+Public console scripts are defined in `pyproject.toml`.
+
+### Dataset Generation
+
+- [x] `dingo_generate_dataset`
+  - File: `dingo/gw/dataset/generate_dataset.py`
+  - Stage 1 has started.
+  - Uses Hydra and `instantiate(...)` for some objects already, so it is partly
+    beyond a purely superficial Stage 1 pass.
+  - Remaining concern: compression/SVD logic is still procedural and should be
+    revisited in Stage 3.
+
+- [ ] `dingo_generate_dataset_dag`
+  - File: `dingo/gw/dataset/generate_dataset_dag.py`
+  - Defer Condor/DAG integration to Stage 3.
+  - Stage 3 goal: submit Condor jobs through a Hydra-aware launcher/workflow
+    rather than hand-written temporary YAML chunks and manual submit files.
+  - This may be analogous in spirit to existing Hydra Slurm launcher workflows,
+    but Condor will likely need project-specific integration.
+  - Estimated Stage 1 difficulty: out of scope.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_evaluate_multibanded_domain`
+  - File: `dingo/gw/dataset/evaluate_multibanded_domain.py`
+  - Stage 1: straightforward wrapper around current settings dict.
+  - Stage 3: replace `build_domain`, `build_prior_with_defaults`, and manual
+    waveform-generator branching with Hydra targets.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_build_svd`
+  - File: `dingo/gw/dataset/utils.py`
+  - Stage 1: small CLI migration.
+  - Stage 3: likely becomes part of a Hydra-instantiated compression/SVD
+    workflow.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_merge_datasets`
+  - File: `dingo/gw/dataset/utils.py`
+  - Stage 1: small CLI migration.
+  - Stage 3: probably little to do.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low.
+
+### Noise / ASD Generation
+
+- [ ] `dingo_generate_asd_dataset`
+  - File: `dingo/gw/noise/generate_dataset.py`
+  - Stage 1: replace parser/settings-file handling with Hydra.
+  - Stage 1 should cover local generation only.
+  - Condor/DAG behavior should remain as-is or be deferred.
+  - Estimated Stage 1 difficulty: medium.
+  - Estimated Stage 3 difficulty: medium/high for Condor integration.
+
+- [ ] `dingo_estimate_psds`
+  - File: `dingo/gw/noise/asd_estimation.py`
+  - Stage 1: straightforward wrapper around current dict settings.
+  - Stage 3: domain construction and PSD-estimation settings could become more
+    declarative.
+  - Estimated Stage 1 difficulty: low/medium.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_generate_synthetic_asd_dataset`
+  - File: `dingo/gw/noise/synthetic/generate_dataset.py`
+  - Stage 1: straightforward config migration.
+  - Stage 3: parameterization and sampling could become separate config groups.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_merge_asd_datasets`
+  - File: `dingo/gw/noise/utils.py`
+  - Stage 1: small CLI migration.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low.
+
+### Training
+
+- [ ] `dingo_train`
+  - File: `dingo/gw/training/train_pipeline.py`
+  - Stage 1: replace parser with Hydra, while still converting to the old
+    `train_settings` and `local_settings` dicts.
+  - Watch out:
+    - New training versus checkpoint resume are currently mutually exclusive CLI
+      modes.
+    - `local` settings are popped out and saved separately as
+      `local_settings.yaml`.
+    - Train settings are saved into model metadata.
+    - Several progress messages still use `print`.
+  - Estimated Stage 1 difficulty: medium/high.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_train_condor`
+  - File: `dingo/gw/training/train_pipeline_condor.py`
+  - Defer to Stage 3 with the rest of Condor integration.
+  - Goal: a Hydra-aware Condor submission path, ideally with a clean
+    single-command link from config composition to job submission.
+  - Watch out:
+    - Reads/writes `local_settings.yaml`.
+    - Creates Condor submit files and mutates local settings.
+  - Estimated Stage 1 difficulty: out of scope.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_append_training_stage`
+  - File: `dingo/gw/training/utils.py`
+  - Stage 1: small Hydra CLI.
+  - Watch out: modifies model metadata/training-stage settings.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low/medium.
+
+### Pipe / Inference
+
+The `dingo_pipe*` family is the largest migration area because it is based on
+`BilbyArgParser` / `configargparse` and has bilby_pipe-style conventions.
+
+- [ ] `dingo_pipe`
+  - File: `dingo/pipe/main.py`
+  - Uses parser from `dingo/pipe/parser.py`.
+  - Estimated Stage 1 difficulty: high.
+  - Estimated Stage 3 difficulty: very high.
+
+- [ ] `dingo_pipe_generation`
+  - File: `dingo/pipe/data_generation.py`
+  - Estimated Stage 1 difficulty: high.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_pipe_sampling`
+  - File: `dingo/pipe/sampling.py`
+  - Estimated Stage 1 difficulty: high.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_pipe_importance_sampling`
+  - File: `dingo/pipe/importance_sampling.py`
+  - Estimated Stage 1 difficulty: high.
+  - Estimated Stage 3 difficulty: high.
+
+- [ ] `dingo_pipe_plot`
+  - File: `dingo/pipe/plot.py`
+  - Estimated Stage 1 difficulty: medium/high.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_pipe_pp_test`
+  - File: `dingo/pipe/pp_test.py`
+  - Estimated Stage 1 difficulty: low/medium.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] `dingo_result`
+  - File: `dingo/pipe/dingo_result.py`
+  - Small CLI, but connected to pipe output conventions.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low.
+
+### Other Utilities
+
+- [ ] `dingo_pt_to_hdf5`
+  - File: `dingo/core/utils/pt_to_hdf5.py`
+  - Small argparse CLI.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low.
+
+- [ ] `dingo_ls`
+  - File: `dingo/gw/ls_cli.py`
+  - Inspection tool rather than a config-driven workflow.
+  - Could stay argparse for a while, but migrate for consistency if Stage 1 aims
+    to eliminate all public parsers.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: low.
+
+- [ ] Importance-weight generation
+  - File: `dingo/gw/importance_sampling/importance_weights.py`
+  - Not currently listed in `project.scripts`, but it is a config-driven CLI.
+  - Estimated Stage 1 difficulty: low/medium.
+  - Estimated Stage 3 difficulty: medium.
+
+- [ ] Unconditional density estimation
+  - File: `dingo/core/density/unconditional_density_estimation.py`
+  - Not currently listed in `project.scripts`, but has an argparse settings CLI.
+  - Estimated Stage 1 difficulty: low.
+  - Estimated Stage 3 difficulty: medium.
+
+## Stage 2 Config Group Candidates
+
+- `domain`
+- `waveform_generator`
+- `intrinsic_prior`
+- `extrinsic_prior`
+- `inference_parameters`
+- `dataset`
+- `asd_dataset`
+- `noise`
+- `model/posterior`
+- `model/embedding`
+- `training/stage`
+- `training/local`
+- `condor` (Stage 3 only)
+- `pipe/event`
+- `pipe/sampler`
+- `importance_sampling`
+- `defaults`
+- `defaults/prior`
+- `defaults/recovery`
+- `defaults/nde`
+
+## Stage 3 Refactor Hotspots
+
+- [ ] Move Python-defined defaults into Hydra default configs
+  - Goal: make default choices visible, overridable, and composable from the
+    config tree rather than hidden in module globals or helper functions.
+  - Related goal: make configs complete. If an option affects behavior, it
+    should appear in a default config somewhere, even if most examples inherit
+    it unchanged.
+  - Likely config groups:
+    - `defaults/prior/intrinsic`
+    - `defaults/prior/extrinsic`
+    - `defaults/inference_parameters`
+    - `defaults/density_recovery`
+    - `defaults/importance_sampling`
+    - `defaults/nde`
+    - `defaults/training`
+  - Keep a deliberate backward-compatibility story for old saved metadata that
+    uses strings such as `default`.
+
+- [ ] Audit implicit constructor/function defaults and expose them in config
+  - Example configs currently omit many options because Python functions and
+    constructors provide defaults.
+  - Stage 3 should decide which defaults are part of the public configuration
+    surface and place them in default configs.
+  - This is different from Stage 1, where copied configs may remain incomplete
+    as long as old code handles them.
+  - Candidate areas:
+    - waveform generator optional kwargs such as `f_start`, `mode_list`,
+      `spin_conversion_phase`, and interface-specific options
+    - dataset compression/SVD/whitening options
+    - training local/runtime/checkpoint options
+    - optimizer and scheduler defaults
+    - transform defaults in the training data pipeline
+    - pipe/inference defaults currently supplied by parser defaults
+    - ASD/noise generation defaults such as frequency range, channels, and
+      Condor settings
+
+- [ ] `dingo/gw/domains/build_domain.py`
+  - Replace `type`-based dispatch with Hydra instantiation.
+  - Keep metadata/backward compatibility in mind.
+
+- [ ] `dingo/gw/prior.py`
+  - Replace `default` string handling and prior builders with explicit config
+   targets or reusable config groups.
+  - Current code-defined defaults:
+    - `default_intrinsic_dict`
+    - `default_extrinsic_dict`
+    - `default_inference_parameters`
+
+- [ ] Waveform generator construction
+  - Replace `new_interface` flags and manual class branching with `_target_`.
+
+- [ ] Compression/SVD workflow
+  - Harder than plain instantiation because SVD may be loaded from file or
+    trained from generated waveforms.
+  - Possible long-term design: compression is an ordered transform list, with
+    SVD handled by a project-specific factory target.
+
+- [ ] `dingo/gw/training/train_builders.py`
+  - Large procedural transform pipeline.
+  - Prime candidate for Hydra targets, but many transforms need runtime objects:
+    domain, detectors, ASD datasets, standardization, priors, reference time.
+
+- [ ] `dingo/core/posterior_models/build_model.py`
+  - Replace `posterior_model_type` dispatch with `_target_`, while preserving
+    saved-model compatibility.
+
+- [ ] `dingo/core/utils/torchutils.py`
+  - Replace optimizer/scheduler `type` switches with `_target_` configs.
+
+- [ ] Condor integration as a Hydra workflow/launcher
+  - Keep Condor-specific migration out of Stage 1.
+  - Stage 3 goal: compose the job config with Hydra and submit Condor jobs from
+    that composed config.
+  - Desired user experience: a single Hydra-driven command can configure and
+    submit the Condor job.
+  - Existing Hydra Slurm launchers may be useful as conceptual references, but
+    HTCondor integration will likely require project-specific code.
+  - Affected areas include:
+    - `dingo/gw/dataset/generate_dataset_dag.py`
+    - `dingo/gw/noise/generate_dataset.py`
+    - `dingo/gw/training/train_pipeline_condor.py`
+
+- [ ] `dingo/pipe/default_settings.py`
+  - Move `DENSITY_RECOVERY_SETTINGS` and `IMPORTANCE_SAMPLING_SETTINGS` to
+    Hydra default configs.
+  - Current named defaults:
+    - `ProxyRecoveryDefault`
+    - `PhaseRecoveryDefault`
+    - `MultibandingDefault`
+
+- [ ] `dingo/core/density/nde_settings.py`
+  - Move `get_default_nde_settings_3d(...)` and `DEFAULT_NDE_SETTINGS_2D` into
+    Hydra configs.
+  - These can likely become reusable `defaults/nde` entries plus overrides for
+    dimension/device/parameters.
+
+- [ ] `dingo/pipe/parser.py`
+  - Biggest migration knot. This is both a CLI parser and a compatibility layer
+    with bilby_pipe-style settings.
+
+## Suggested Order
+
+1. Finish Stage 1 for standalone dataset/noise utilities.
+2. Do Stage 1 for training.
+3. Decide whether `dingo_pipe` should be a compatibility wrapper around Hydra
+   configs or a full Hydra CLI.
+4. Do Stage 2 grouping for dataset generation, ASD generation, and training
+   before touching pipe internals.
+5. Start Stage 3 with the small builder switches:
+   domain, prior, waveform generator, optimizer, scheduler.
+6. Design Condor integration as a Stage 3 Hydra workflow/launcher problem.
+7. Tackle training transforms and model construction.
+8. Leave pipe/parser migration until the rest of the config story is stable.
diff --git a/configs/examples/generate_dataset.yaml b/configs/examples/generate_dataset.yaml
new file mode 100644
index 000000000..6ac937e0e
--- /dev/null
+++ b/configs/examples/generate_dataset.yaml
@@ -0,0 +1,42 @@
+# settings for domain of waveforms
+domain:
+  type: UniformFrequencyDomain
+  f_min: 20.0
+  f_max: 1024.0
+  delta_f: 0.25  # Expressions like 1.0/8.0 would require eval and are not supported
+
+# settings for waveform generator
+waveform_generator:
+  approximant: IMRPhenomD
+  f_ref: 20.0
+  # f_start: 15.0  # Optional setting useful for EOB waveforms. Overrides f_min when generating waveforms.
+
+# Dataset only samples over intrinsic parameters. Extrinsic parameters are chosen at train time.
+intrinsic_prior:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: default
+  chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
+  chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
+  theta_jn: default
+  # Reference values for fixed (extrinsic) parameters. These are needed to generate a waveform.
+  luminosity_distance: 100.0  # Mpc
+  geocent_time: 0.0  # s
+
+# Dataset size
+num_samples: 10000
+
+compression: null
+
+num_processes: 1
+
+out_file: waveform_dataset.hdf5
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: generate_dataset
+    chdir: true
diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index 0b4fda8a0..b35cf49c3 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -9,15 +9,17 @@
 import numpy as np
 import pandas as pd
 from bilby.gw.prior import BBHPriorDict
-from hydra.utils import instantiate
 from omegaconf import DictConfig, OmegaConf
 from threadpoolctl import threadpool_limits
 from torchvision.transforms import Compose
 
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
+from dingo.gw.domains import build_domain
+from dingo.gw.prior import build_prior_with_defaults
 from dingo.gw.SVD import ApplySVD, SVDBasis
 from dingo.gw.transforms import WhitenFixedASD
 from dingo.gw.waveform_generator import (
+    NewInterfaceWaveformGenerator,
     WaveformGenerator,
     generate_waveforms_parallel,
 )
@@ -139,33 +141,43 @@ def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int):
     return basis, n_train, n_test
 
 
-def _dataset_settings(cfg: DictConfig) -> Dict:
+def _settings_from_config(cfg: DictConfig) -> Dict:
     settings = OmegaConf.to_container(cfg, resolve=True)
     settings.pop("out_file", None)
     settings.pop("num_processes", None)
     return settings
 
 
-def generate_dataset(cfg: DictConfig) -> WaveformDataset:
+def generate_dataset(settings: Dict, num_processes: int) -> WaveformDataset:
     """
     Generate a waveform dataset.
 
     Parameters
     ----------
-    cfg : DictConfig
-        Hydra configuration for dataset generation.
+    settings : dict
+        Dictionary of settings to configure the dataset
+    num_processes : int
 
     Returns
     -------
     A WaveformDataset based on the settings.
     """
 
-    prior = instantiate(cfg.intrinsic_prior, _convert_="all")
-    domain = instantiate(cfg.domain)
-    waveform_generator = instantiate(cfg.waveform_generator, domain=domain)
-    num_processes = cfg.get("num_processes", 1)
+    prior = build_prior_with_defaults(settings["intrinsic_prior"])
+    domain = build_domain(settings["domain"])
+
+    new_interface_flag = settings["waveform_generator"].get("new_interface", False)
+    if new_interface_flag:
+        waveform_generator = NewInterfaceWaveformGenerator(
+            domain=domain,
+            **settings["waveform_generator"],
+        )
+    else:
+        waveform_generator = WaveformGenerator(
+            domain=domain,
+            **settings["waveform_generator"],
+        )
 
-    settings = _dataset_settings(cfg)
     dataset_dict = {"settings": settings}
 
     if settings.get("compression", None) is not None:
@@ -243,13 +255,14 @@ def generate_dataset(cfg: DictConfig) -> WaveformDataset:
     dataset_dict["parameters"] = parameters
     dataset_dict["polarizations"] = polarizations
 
+    dataset_dict[settings["num_samples"]] = len(parameters)
     dataset = WaveformDataset(dictionary=dataset_dict)
     return dataset
 
 
 @hydra.main(
     version_base="1.3",
-    config_path="../../../configs/gw/dataset",
+    config_path="../../../configs/examples",
     config_name="generate_dataset",
 )
 def main(cfg: DictConfig) -> None:
@@ -259,7 +272,8 @@ def main(cfg: DictConfig) -> None:
             f"dataset generation: can not create {out_file}: "
             f"{Path(out_file).parent} does not exist"
         )
-    dataset = generate_dataset(cfg)
+    settings = _settings_from_config(cfg)
+    dataset = generate_dataset(settings, cfg.num_processes)
     dataset.to_file(str(out_file))
 
 
diff --git a/pyproject.toml b/pyproject.toml
index 7eef484df..e3341fc16 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -33,6 +33,7 @@ dependencies = [
     "glasflow",
     "gwpy",
     "h5py",
+    "hydra-core",
     "lalsuite>=7.15",
     "matplotlib",
     "numpy",
diff --git a/uv.lock b/uv.lock
index 3b604a4a2..fbdef97b1 100644
--- a/uv.lock
+++ b/uv.lock
@@ -28,6 +28,12 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/78/b6/6307fbef88d9b5ee7421e68d78a9f162e0da4900bc5f5793f6d3d0e34fb8/annotated_types-0.7.0-py3-none-any.whl", hash = "sha256:1f02e8b43a8fbbc3f3e0d4f0f4bfc8131bcb4eebe8849b8e5c773f3a1c582a53", size = 13643, upload-time = "2024-05-20T21:33:24.1Z" },
 ]
 
+[[package]]
+name = "antlr4-python3-runtime"
+version = "4.9.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/3e/38/7859ff46355f76f8d19459005ca000b6e7012f2f1ca597746cbcd1fbfe5e/antlr4-python3-runtime-4.9.3.tar.gz", hash = "sha256:f224469b4168294902bb1efa80a8bf7855f24c99aef99cbefc1bcd3cce77881b", size = 117034, upload-time = "2021-11-06T17:52:23.524Z" }
+
 [[package]]
 name = "appnope"
 version = "0.1.4"
@@ -78,11 +84,11 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy-iers-data", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "packaging", marker = "python_full_version < '3.11'" },
-    { name = "pyerfa", marker = "python_full_version < '3.11'" },
-    { name = "pyyaml", marker = "python_full_version < '3.11'" },
+    { name = "astropy-iers-data" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "packaging" },
+    { name = "pyerfa" },
+    { name = "pyyaml" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/8a/f8/9c6675ab4c646b95aae2762d108f6be4504033d91bd50da21daa62cab5ce/astropy-6.1.7.tar.gz", hash = "sha256:a405ac186306b6cb152e6df2f7444ab8bd764e4127d7519da1b3ae4dd65357ef", size = 7063411, upload-time = "2024-11-22T21:22:34.373Z" }
 wheels = [
@@ -127,11 +133,11 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy-iers-data", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "packaging", marker = "python_full_version >= '3.11'" },
-    { name = "pyerfa", marker = "python_full_version >= '3.11'" },
-    { name = "pyyaml", marker = "python_full_version >= '3.11'" },
+    { name = "astropy-iers-data" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "packaging" },
+    { name = "pyerfa" },
+    { name = "pyyaml" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/7b/92/2dce2d48347efc3346d08ca7995b152d242ebd170c571f7c9346468d8427/astropy-7.2.0.tar.gz", hash = "sha256:ae48bc26b1feaeb603cd94bd1fa1aa39137a115fe931b7f13787ab420e8c3070", size = 7057774, upload-time = "2025-11-25T22:36:41.916Z" }
 wheels = [
@@ -265,24 +271,24 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "bilby", extra = ["gw"], marker = "python_full_version < '3.11'" },
-    { name = "configargparse", marker = "python_full_version < '3.11'" },
-    { name = "corner", marker = "python_full_version < '3.11'" },
-    { name = "dynesty", marker = "python_full_version < '3.11'" },
-    { name = "future", marker = "python_full_version < '3.11'" },
-    { name = "gwosc", marker = "python_full_version < '3.11'" },
-    { name = "gwpy", version = "3.0.13", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "jinja2", marker = "python_full_version < '3.11'" },
-    { name = "ligo-gracedb", marker = "python_full_version < '3.11'" },
-    { name = "matplotlib", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "plotly", marker = "python_full_version < '3.11'" },
-    { name = "pycondor", marker = "python_full_version < '3.11'" },
-    { name = "python-ligo-lw", marker = "python_full_version < '3.11'" },
-    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "seaborn", marker = "python_full_version < '3.11'" },
-    { name = "tqdm", marker = "python_full_version < '3.11'" },
+    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" } },
+    { name = "bilby", extra = ["gw"] },
+    { name = "configargparse" },
+    { name = "corner" },
+    { name = "dynesty" },
+    { name = "future" },
+    { name = "gwosc" },
+    { name = "gwpy", version = "3.0.13", source = { registry = "https://pypi.org/simple" } },
+    { name = "jinja2" },
+    { name = "ligo-gracedb" },
+    { name = "matplotlib" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "plotly" },
+    { name = "pycondor" },
+    { name = "python-ligo-lw" },
+    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "seaborn" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/11/89/231ec5533fd46c40b080bb1dd7aa8947f69b6ca65259c048804e2ae9aa6f/bilby_pipe-1.8.0.tar.gz", hash = "sha256:b6155b82f2dc3afbd5de69598641edc716b6b8331c902ef35c9a3a2d547dc83a", size = 57085085, upload-time = "2025-12-20T15:58:56.843Z" }
 wheels = [
@@ -300,24 +306,24 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "bilby", extra = ["gw"], marker = "python_full_version >= '3.11'" },
-    { name = "configargparse", marker = "python_full_version >= '3.11'" },
-    { name = "corner", marker = "python_full_version >= '3.11'" },
-    { name = "dynesty", marker = "python_full_version >= '3.11'" },
-    { name = "future", marker = "python_full_version >= '3.11'" },
-    { name = "gwosc", marker = "python_full_version >= '3.11'" },
-    { name = "gwpy", version = "4.0.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "igwn-ligolw", version = "2.1.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "jinja2", marker = "python_full_version >= '3.11'" },
-    { name = "ligo-gracedb", marker = "python_full_version >= '3.11'" },
-    { name = "matplotlib", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "plotly", marker = "python_full_version >= '3.11'" },
-    { name = "pycondor", marker = "python_full_version >= '3.11'" },
-    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "seaborn", marker = "python_full_version >= '3.11'" },
-    { name = "tqdm", marker = "python_full_version >= '3.11'" },
+    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "bilby", extra = ["gw"] },
+    { name = "configargparse" },
+    { name = "corner" },
+    { name = "dynesty" },
+    { name = "future" },
+    { name = "gwosc" },
+    { name = "gwpy", version = "4.0.1", source = { registry = "https://pypi.org/simple" } },
+    { name = "igwn-ligolw", version = "2.1.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "jinja2" },
+    { name = "ligo-gracedb" },
+    { name = "matplotlib" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "plotly" },
+    { name = "pycondor" },
+    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "seaborn" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/e3/41/66780cb4fdb3d008a593a175e17bec61fd2f88f8609cd68860ec4a308302/bilby_pipe-1.8.1.tar.gz", hash = "sha256:1b287f2db4756770502d10e05c12b941ea2837f1f96e273ff0fe2949a407fab0", size = 57085114, upload-time = "2026-04-07T17:53:20.489Z" }
 wheels = [
@@ -562,7 +568,7 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/66/54/eb9bfc647b19f2009dd5c7f5ec51c4e6ca831725f1aea7a993034f483147/contourpy-1.3.2.tar.gz", hash = "sha256:b6945942715a034c671b7fc54f9588126b0b8bf23db2696e3ca8328f3ff0ab54", size = 13466130, upload-time = "2025-04-15T17:47:53.79Z" }
 wheels = [
@@ -635,7 +641,7 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/58/01/1253e6698a07380cd31a736d248a3f2a50a7c88779a1813da27503cadc2a/contourpy-1.3.3.tar.gz", hash = "sha256:083e12155b210502d0bca491432bb04d56dc3432f95a979b429f2848c3dbe880", size = 13466174, upload-time = "2025-07-26T12:03:12.549Z" }
 wheels = [
@@ -864,6 +870,7 @@ dependencies = [
     { name = "gwpy", version = "3.0.13", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
     { name = "gwpy", version = "4.0.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
     { name = "h5py" },
+    { name = "hydra-core" },
     { name = "lalsuite" },
     { name = "matplotlib" },
     { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
@@ -948,6 +955,7 @@ requires-dist = [
     { name = "gwpy" },
     { name = "h5py" },
     { name = "htcondor", marker = "extra == 'dev'" },
+    { name = "hydra-core" },
     { name = "ipykernel", marker = "extra == 'dev'" },
     { name = "lalsuite", specifier = ">=7.15" },
     { name = "linkify-it-py", marker = "extra == 'dev'" },
@@ -1058,7 +1066,7 @@ name = "exceptiongroup"
 version = "1.3.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+    { name = "typing-extensions" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/50/79/66800aadf48771f6b62f7eb014e352e5d06856655206165d775e675a02c9/exceptiongroup-1.3.1.tar.gz", hash = "sha256:8b412432c6055b0b7d14c310000ae93352ed6754f70fa8f7c34141f91c4e3219", size = 30371, upload-time = "2025-11-21T23:01:54.787Z" }
 wheels = [
@@ -1276,21 +1284,21 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "dateparser", marker = "python_full_version < '3.11'" },
-    { name = "dqsegdb2", marker = "python_full_version < '3.11'" },
-    { name = "gwdatafind", marker = "python_full_version < '3.11'" },
-    { name = "gwosc", marker = "python_full_version < '3.11'" },
-    { name = "h5py", marker = "python_full_version < '3.11'" },
-    { name = "igwn-segments", marker = "python_full_version < '3.11'" },
-    { name = "ligotimegps", version = "2.0.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "matplotlib", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "packaging", marker = "python_full_version < '3.11'" },
-    { name = "python-dateutil", marker = "python_full_version < '3.11'" },
-    { name = "requests", marker = "python_full_version < '3.11'" },
-    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "tqdm", marker = "python_full_version < '3.11'" },
+    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" } },
+    { name = "dateparser" },
+    { name = "dqsegdb2" },
+    { name = "gwdatafind" },
+    { name = "gwosc" },
+    { name = "h5py" },
+    { name = "igwn-segments" },
+    { name = "ligotimegps", version = "2.0.1", source = { registry = "https://pypi.org/simple" } },
+    { name = "matplotlib" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "packaging" },
+    { name = "python-dateutil" },
+    { name = "requests" },
+    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/ff/5e/a210178c2ca8ac3ff2df25634f960077a1b537e9187321c27a101bb99d50/gwpy-3.0.13.tar.gz", hash = "sha256:59a2dc90a9c2edde3efae090303c0f8afeb6c644dae6c8fdd4d92bf0799374b1", size = 1542843, upload-time = "2025-07-02T09:30:47.183Z" }
 wheels = [
@@ -1308,21 +1316,21 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "dateparser", marker = "python_full_version >= '3.11'" },
-    { name = "dqsegdb2", marker = "python_full_version >= '3.11'" },
-    { name = "gwdatafind", marker = "python_full_version >= '3.11'" },
-    { name = "gwosc", marker = "python_full_version >= '3.11'" },
-    { name = "h5py", marker = "python_full_version >= '3.11'" },
-    { name = "igwn-segments", marker = "python_full_version >= '3.11'" },
-    { name = "ligotimegps", version = "2.1.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "matplotlib", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "packaging", marker = "python_full_version >= '3.11'" },
-    { name = "python-dateutil", marker = "python_full_version >= '3.11'" },
-    { name = "requests", marker = "python_full_version >= '3.11'" },
-    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "tqdm", marker = "python_full_version >= '3.11'" },
+    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "dateparser" },
+    { name = "dqsegdb2" },
+    { name = "gwdatafind" },
+    { name = "gwosc" },
+    { name = "h5py" },
+    { name = "igwn-segments" },
+    { name = "ligotimegps", version = "2.1.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "matplotlib" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "packaging" },
+    { name = "python-dateutil" },
+    { name = "requests" },
+    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/92/71/1e652734a0e8f23e0f9dbf5df5698e66fa49aef4446d2c6c9eb914921192/gwpy-4.0.1.tar.gz", hash = "sha256:477aa69bd40506bfb0df5ec779d9dd6fef562a3bfabe58a52b297e25c57f61f0", size = 1615849, upload-time = "2026-02-03T10:14:59.045Z" }
 wheels = [
@@ -1374,17 +1382,31 @@ wheels = [
 
 [[package]]
 name = "htcondor"
-version = "25.7.2"
+version = "25.11.0"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/da/7c/b88e13d03ab47c930da456bdfbe0428e9098fe8341c099e4f200f44c6830/htcondor-25.7.2-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:2c30fff3d704437786ce5526d6a8e444c1246b2996d264a07254c31415488562", size = 50208679, upload-time = "2026-03-12T15:42:42.497Z" },
-    { url = "https://files.pythonhosted.org/packages/ba/0e/87386e8028934d502d63b8314d830aaa6b9fe209a2b3173a423fb7ab54a3/htcondor-25.7.2-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:ac6bc73c338828b73f44fdcd4ff5407f70810314bd1e160a99fcd2c3f2e54f34", size = 51634164, upload-time = "2026-03-12T15:42:46.353Z" },
-    { url = "https://files.pythonhosted.org/packages/bf/fa/027cd625887cf4e37ddff4c33dc9e9c5e9ef18d620b14babc3da98f175c3/htcondor-25.7.2-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:60fb891fae4680743b5f72108ec29020094edbf7fb3214a49055bcd79ac7fd00", size = 50208821, upload-time = "2026-03-12T15:42:50.351Z" },
-    { url = "https://files.pythonhosted.org/packages/c5/06/6a48f90656c2d252c79449821f694612d002a6ad37d822778df51a2d3bef/htcondor-25.7.2-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:c020d64225daf8bb3c765fab933859dfa73e899485fcd57767c46d736ab9dfe5", size = 51640592, upload-time = "2026-03-12T15:42:54.822Z" },
-    { url = "https://files.pythonhosted.org/packages/a5/82/536f95681b0a1bea5f3709917dba270133e93b14626cff36e80959bea13a/htcondor-25.7.2-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:94af431820294fea9b28abae7298ca3d7d52ab584f2c79327d27eff03f8cc5ae", size = 50209472, upload-time = "2026-03-12T15:42:58.991Z" },
-    { url = "https://files.pythonhosted.org/packages/90/ef/4725f304f0e718cb0f0fca00fdf44a21ff3d42116553a62ee9026a52ed43/htcondor-25.7.2-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:d462cf68b7ca39ee91923c3716ee9e0c46713a153232b5f7b68643cb91448621", size = 51631016, upload-time = "2026-03-12T15:43:04.048Z" },
-    { url = "https://files.pythonhosted.org/packages/88/03/afef2f6d0d3cacefa2c2a97736439ad44c3b77148d77caa20bbfa507ac16/htcondor-25.7.2-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:5097700ec9c609c752f2e76665031b9925633cff2cba6fe1db2d2ef270e88dcd", size = 50212296, upload-time = "2026-03-12T15:43:08.521Z" },
-    { url = "https://files.pythonhosted.org/packages/d8/94/350a5e0c412095666cf1e771969e226072ab6233ebceabd5994337aad726/htcondor-25.7.2-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:306bf1f714299341b556d46b3ea662822d8ef58780b450177bce5b720ce24e3b", size = 51645252, upload-time = "2026-03-12T15:43:13.913Z" },
+    { url = "https://files.pythonhosted.org/packages/94/05/4582995aa615d0ac0e2b20379a3cfd656b5b67147fc71eb028331ebc4569/htcondor-25.11.0-cp310-cp310-manylinux_2_28_aarch64.whl", hash = "sha256:9580aea503bf0b009d0831876028f8e6aaec010cc2526789a1b388790d27efa5", size = 51111733, upload-time = "2026-06-11T14:29:16.357Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/96/935019a84d2cb1c74bcd94ab92fea87c83f812dd008cf14271c3179c40ff/htcondor-25.11.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:9a3dfc83e3b498ccbee7198eb570508a6b8e614aaf7dbbe87906ad8a69dfbc93", size = 52554834, upload-time = "2026-06-11T14:29:19.451Z" },
+    { url = "https://files.pythonhosted.org/packages/02/cf/052453b10b58673ebbe55d88fe84a307c42c30a39b949714fb123ecea962/htcondor-25.11.0-cp311-cp311-manylinux_2_28_aarch64.whl", hash = "sha256:6c4e9a74d1eca96cfe7fd83ac38b1f6e79310ff90450f480f2e9d508b5f8c16a", size = 51111931, upload-time = "2026-06-11T14:29:22.461Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/fd/9ab2e532a977a809a7a103d5b3d547b9d772f6ca745d516f3f5269ab12ae/htcondor-25.11.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:77f0ae0bf51e2f5227d5ee18804e642ed7643472d59de631e2a164a6643e7db8", size = 52554719, upload-time = "2026-06-11T14:29:25.355Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/eb/f2f8387f16be10f6bbd92a4b2b2d44c8b29c732bedf5bbe60e14b6e0a30e/htcondor-25.11.0-cp312-cp312-manylinux_2_28_aarch64.whl", hash = "sha256:5f5a2951acb37bc7554af8ee377fa5ed507eca4e31c219643019b59150ff325b", size = 51114270, upload-time = "2026-06-11T14:29:28.452Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/01/219e1047d40cc2254a5a8b791c5c7d26bbf336f7f9b3ebdad57f4d762b4a/htcondor-25.11.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:32b95deb53bb0129983d9e341b22d96241da830f6acdf61c4594815df2c38193", size = 52552026, upload-time = "2026-06-11T14:29:31.297Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/38/da64ab166e271fa3088f573fcda7e4e58817bd9e209f2c7c41e227be95d7/htcondor-25.11.0-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:8cd7efde67d42cbbda3e7f2a564c4c0b27ca7ee60eb8f0482c538628a4176def", size = 51125765, upload-time = "2026-06-11T14:29:34.152Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/f2/b1b3133d3401f795b1975f093ad95e6d2d8fe8c0df0429e321e4e5832573/htcondor-25.11.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:46252c51c9aebf0c9412c52b2c7602ba22734a8157ae9c2983e9dce2547f1d0b", size = 52563721, upload-time = "2026-06-11T14:29:36.875Z" },
+]
+
+[[package]]
+name = "hydra-core"
+version = "1.3.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "antlr4-python3-runtime" },
+    { name = "omegaconf" },
+    { name = "packaging" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/0d/0b/7c0d941311aadc6479ec01767edba9c8a07db1452685de3567ed3058d0c9/hydra_core-1.3.3.tar.gz", hash = "sha256:b7477ee21f08b62f71bf0126d44695c048dc7e9c0cc79e2d593b707cb1e44048", size = 3262532, upload-time = "2026-06-11T05:54:26.835Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e5/57/4e39f85347f77144d2ad12e87d5df8fb8f17023f9bd9e8c6e903a128382c/hydra_core-1.3.3-py3-none-any.whl", hash = "sha256:cf349fc393f486f250e5825592c3d0a50c0af3effd726cf8dd5b637a7cb464e3", size = 154706, upload-time = "2026-06-11T05:54:24.917Z" },
 ]
 
 [[package]]
@@ -1420,11 +1442,11 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "igwn-segments", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "python-dateutil", marker = "python_full_version < '3.11'" },
-    { name = "pyyaml", marker = "python_full_version < '3.11'" },
-    { name = "tqdm", marker = "python_full_version < '3.11'" },
+    { name = "igwn-segments" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "python-dateutil" },
+    { name = "pyyaml" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/1b/f4/ec600abf6891ffc203547b69949be5b34b03bb705433da385a562b93a796/igwn_ligolw-2.0.1.tar.gz", hash = "sha256:e5d740830d497dfcf13d1101b284a1f8bed98d5bd8164a0c236ab4a97eddd2fb", size = 2333107, upload-time = "2025-04-08T12:33:42.741Z" }
 wheels = [
@@ -1457,11 +1479,11 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "igwn-segments", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "python-dateutil", marker = "python_full_version >= '3.11'" },
-    { name = "pyyaml", marker = "python_full_version >= '3.11'" },
-    { name = "tqdm", marker = "python_full_version >= '3.11'" },
+    { name = "igwn-segments" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "python-dateutil" },
+    { name = "pyyaml" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/87/db/4c9420b13b1fd874b4cbc3beb74c96c2d4e17a5fb9c2ead91cdca7b2cfc9/igwn_ligolw-2.1.0.tar.gz", hash = "sha256:02a7201445d5dba8179a28f097b1ca9839a6564a5503201ee922dcbc9aa31c9d", size = 2337001, upload-time = "2025-05-02T12:50:34.755Z" }
 wheels = [
@@ -1596,17 +1618,17 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "colorama", marker = "python_full_version < '3.11' and sys_platform == 'win32'" },
-    { name = "decorator", marker = "python_full_version < '3.11'" },
-    { name = "exceptiongroup", marker = "python_full_version < '3.11'" },
-    { name = "jedi", marker = "python_full_version < '3.11'" },
-    { name = "matplotlib-inline", marker = "python_full_version < '3.11'" },
-    { name = "pexpect", marker = "python_full_version < '3.11' and sys_platform != 'emscripten' and sys_platform != 'win32'" },
-    { name = "prompt-toolkit", marker = "python_full_version < '3.11'" },
-    { name = "pygments", marker = "python_full_version < '3.11'" },
-    { name = "stack-data", marker = "python_full_version < '3.11'" },
-    { name = "traitlets", marker = "python_full_version < '3.11'" },
-    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "decorator" },
+    { name = "exceptiongroup" },
+    { name = "jedi" },
+    { name = "matplotlib-inline" },
+    { name = "pexpect", marker = "sys_platform != 'emscripten' and sys_platform != 'win32'" },
+    { name = "prompt-toolkit" },
+    { name = "pygments" },
+    { name = "stack-data" },
+    { name = "traitlets" },
+    { name = "typing-extensions" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/85/31/10ac88f3357fc276dc8a64e8880c82e80e7459326ae1d0a211b40abf6665/ipython-8.37.0.tar.gz", hash = "sha256:ca815841e1a41a1e6b73a0b08f3038af9b2252564d01fc405356d34033012216", size = 5606088, upload-time = "2025-05-31T16:39:09.613Z" }
 wheels = [
@@ -1624,17 +1646,17 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "colorama", marker = "python_full_version >= '3.11' and sys_platform == 'win32'" },
-    { name = "decorator", marker = "python_full_version >= '3.11'" },
-    { name = "ipython-pygments-lexers", marker = "python_full_version >= '3.11'" },
-    { name = "jedi", marker = "python_full_version >= '3.11'" },
-    { name = "matplotlib-inline", marker = "python_full_version >= '3.11'" },
-    { name = "pexpect", marker = "python_full_version >= '3.11' and sys_platform != 'emscripten' and sys_platform != 'win32'" },
-    { name = "prompt-toolkit", marker = "python_full_version >= '3.11'" },
-    { name = "pygments", marker = "python_full_version >= '3.11'" },
-    { name = "stack-data", marker = "python_full_version >= '3.11'" },
-    { name = "traitlets", marker = "python_full_version >= '3.11'" },
-    { name = "typing-extensions", marker = "python_full_version == '3.11.*'" },
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "decorator" },
+    { name = "ipython-pygments-lexers" },
+    { name = "jedi" },
+    { name = "matplotlib-inline" },
+    { name = "pexpect", marker = "sys_platform != 'emscripten' and sys_platform != 'win32'" },
+    { name = "prompt-toolkit" },
+    { name = "pygments" },
+    { name = "stack-data" },
+    { name = "traitlets" },
+    { name = "typing-extensions", marker = "python_full_version < '3.12'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/12/51/a703c030f4928646d390b4971af4938a1b10c9dfce694f0d99a0bb073cb2/ipython-9.8.0.tar.gz", hash = "sha256:8e4ce129a627eb9dd221c41b1d2cdaed4ef7c9da8c17c63f6f578fe231141f83", size = 4424940, upload-time = "2025-12-03T10:18:24.353Z" }
 wheels = [
@@ -1646,7 +1668,7 @@ name = "ipython-pygments-lexers"
 version = "1.1.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "pygments", marker = "python_full_version >= '3.11'" },
+    { name = "pygments" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/ef/4c/5dd1d8af08107f88c7f741ead7a40854b8ac24ddf9ae850afbcf698aa552/ipython_pygments_lexers-1.1.1.tar.gz", hash = "sha256:09c0138009e56b6854f9535736f4171d855c8c08a563a0dcd8022f78355c7e81", size = 8393, upload-time = "2025-01-17T11:24:34.505Z" }
 wheels = [
@@ -1944,7 +1966,7 @@ name = "ligo-segments"
 version = "1.4.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "six", marker = "python_full_version < '3.11'" },
+    { name = "six" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/81/60/8de5c89e4e5fc760649cee0c773418ecc920c3dae21ac5656fd3e5e21a9d/ligo-segments-1.4.0.tar.gz", hash = "sha256:e072a844713c5b02efdcaf5bfe4c3a8cd9ef225b08cfd3202a4e185e0f71f5dc", size = 51015, upload-time = "2021-10-18T14:47:39.224Z" }
 wheels = [
@@ -2640,7 +2662,7 @@ name = "numpy-typing-compat"
 version = "20251206.2.3"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/77/83/dd90774d6685664cbe5525645a50c4e6c7454207aee552918790e879137f/numpy_typing_compat-20251206.2.3.tar.gz", hash = "sha256:18e00e0f4f2040fe98574890248848c7c6831a975562794da186cf4f3c90b935", size = 5009, upload-time = "2025-12-06T20:02:04.177Z" }
 wheels = [
@@ -2781,6 +2803,19 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/a2/eb/86626c1bbc2edb86323022371c39aa48df6fd8b0a1647bc274577f72e90b/nvidia_nvtx_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5b17e2001cc0d751a5bc2c6ec6d26ad95913324a4adb86788c944f8ce9ba441f", size = 89954, upload-time = "2025-03-07T01:42:44.131Z" },
 ]
 
+[[package]]
+name = "omegaconf"
+version = "2.3.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "antlr4-python3-runtime" },
+    { name = "pyyaml" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/ce/3d/e4b57b8d9008c6ebe0d5eff901f91d5700cf7bdb8c8863df817463a7fd5e/omegaconf-2.3.1.tar.gz", hash = "sha256:e5e7de64aeebeddaf8e6d3f7a783b32ac2a01c0fbd9c878012caecb891a1f42a", size = 3298472, upload-time = "2026-06-11T05:05:12.885Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a4/0e/152509871bf30df6fc38569f52a2db9b55dd41aae957adae50a053ac7778/omegaconf-2.3.1-py3-none-any.whl", hash = "sha256:3d701d14e9a8828f1edd28bb70b725908b34277cdd72cf7d6a83f94dadc6b6a0", size = 79502, upload-time = "2026-06-11T05:05:09.954Z" },
+]
+
 [[package]]
 name = "optype"
 version = "0.9.3"
@@ -2790,7 +2825,7 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+    { name = "typing-extensions" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/88/3c/9d59b0167458b839273ad0c4fc5f62f787058d8f5aed7f71294963a99471/optype-0.9.3.tar.gz", hash = "sha256:5f09d74127d316053b26971ce441a4df01f3a01943601d3712dd6f34cdfbaf48", size = 96143, upload-time = "2025-03-31T17:00:08.392Z" }
 wheels = [
@@ -2808,7 +2843,7 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "typing-extensions", marker = "python_full_version >= '3.11' and python_full_version < '3.13'" },
+    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/d7/93/6b9e43138ce36fbad134bd1a50460a7bbda61105b5a964e4cf773fe4d845/optype-0.15.0.tar.gz", hash = "sha256:457d6ca9e7da19967ec16d42bdf94e240b33b5d70a56fbbf5b427e5ea39cf41e", size = 99978, upload-time = "2025-12-08T12:32:41.422Z" }
 wheels = [
@@ -2817,8 +2852,8 @@ wheels = [
 
 [package.optional-dependencies]
 numpy = [
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "numpy-typing-compat", marker = "python_full_version >= '3.11'" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "numpy-typing-compat" },
 ]
 
 [[package]]
@@ -3161,26 +3196,26 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "beautifulsoup4", marker = "python_full_version < '3.11'" },
-    { name = "cython", marker = "python_full_version < '3.11'" },
-    { name = "gwdatafind", marker = "python_full_version < '3.11'" },
-    { name = "h5py", marker = "python_full_version < '3.11'" },
-    { name = "igwn-ligolw", version = "2.0.1", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "igwn-segments", marker = "python_full_version < '3.11'" },
-    { name = "jinja2", marker = "python_full_version < '3.11'" },
-    { name = "lalsuite", marker = "python_full_version < '3.11'" },
-    { name = "lscsoft-glue", marker = "python_full_version < '3.11'" },
-    { name = "mako", marker = "python_full_version < '3.11'" },
-    { name = "matplotlib", marker = "python_full_version < '3.11'" },
-    { name = "mpld3", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "pegasus-wms-api", marker = "python_full_version < '3.11'" },
-    { name = "pillow", marker = "python_full_version < '3.11'" },
-    { name = "pykerr", marker = "python_full_version < '3.11'" },
-    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "setuptools", marker = "python_full_version < '3.11'" },
-    { name = "tqdm", marker = "python_full_version < '3.11'" },
+    { name = "astropy", version = "6.1.7", source = { registry = "https://pypi.org/simple" } },
+    { name = "beautifulsoup4" },
+    { name = "cython" },
+    { name = "gwdatafind" },
+    { name = "h5py" },
+    { name = "igwn-ligolw", version = "2.0.1", source = { registry = "https://pypi.org/simple" } },
+    { name = "igwn-segments" },
+    { name = "jinja2" },
+    { name = "lalsuite" },
+    { name = "lscsoft-glue" },
+    { name = "mako" },
+    { name = "matplotlib" },
+    { name = "mpld3" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "pegasus-wms-api" },
+    { name = "pillow" },
+    { name = "pykerr" },
+    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "setuptools" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/fe/9b/834a129763dc752db22eecacc71bfa3fa0b04e1cfd9c7b00a1b417c837ec/PyCBC-2.9.0.tar.gz", hash = "sha256:f9bf981814d6d522ad0c9affadcfd3f3d2ea8422b047d70755b06c03c08e342e", size = 3574557, upload-time = "2025-07-03T13:22:45.353Z" }
 wheels = [
@@ -3206,26 +3241,26 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "beautifulsoup4", marker = "python_full_version >= '3.11'" },
-    { name = "cython", marker = "python_full_version >= '3.11'" },
-    { name = "gwdatafind", marker = "python_full_version >= '3.11'" },
-    { name = "h5py", marker = "python_full_version >= '3.11'" },
-    { name = "igwn-ligolw", version = "2.1.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "igwn-segments", marker = "python_full_version >= '3.11'" },
-    { name = "jinja2", marker = "python_full_version >= '3.11'" },
-    { name = "lalsuite", marker = "python_full_version >= '3.11'" },
-    { name = "lscsoft-glue", marker = "python_full_version >= '3.11'" },
-    { name = "mako", marker = "python_full_version >= '3.11'" },
-    { name = "matplotlib", marker = "python_full_version >= '3.11'" },
-    { name = "mpld3", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "pegasus-wms-api", marker = "python_full_version >= '3.11'" },
-    { name = "pillow", marker = "python_full_version >= '3.11'" },
-    { name = "pykerr", marker = "python_full_version >= '3.11'" },
-    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "setuptools", marker = "python_full_version >= '3.11'" },
-    { name = "tqdm", marker = "python_full_version >= '3.11'" },
+    { name = "astropy", version = "7.2.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "beautifulsoup4" },
+    { name = "cython" },
+    { name = "gwdatafind" },
+    { name = "h5py" },
+    { name = "igwn-ligolw", version = "2.1.0", source = { registry = "https://pypi.org/simple" } },
+    { name = "igwn-segments" },
+    { name = "jinja2" },
+    { name = "lalsuite" },
+    { name = "lscsoft-glue" },
+    { name = "mako" },
+    { name = "matplotlib" },
+    { name = "mpld3" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "pegasus-wms-api" },
+    { name = "pillow" },
+    { name = "pykerr" },
+    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "setuptools" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/a8/3e/bd821e5760ceab12de6607a4a385e78f301951531c2453b21e167484e966/pycbc-2.10.0.tar.gz", hash = "sha256:3c1034e4089d256936c4d80f9c1fa1348c1ff604357a47a4beffbcac2c135edf", size = 3595057, upload-time = "2026-01-07T15:59:25.808Z" }
 wheels = [
@@ -3394,8 +3429,8 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "setuptools", marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "setuptools" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/4b/3f/ee1bc44b080fc1e81d293cd07bed563d254bc1997d63a3b8053804a87dfd/pyfftw-0.15.0.tar.gz", hash = "sha256:2f16b9854a40c8fdd10aa5803b24ddc6ab49f9cd559dbd7f07e7d61aa205c1ca", size = 164003, upload-time = "2024-11-06T16:01:19.293Z" }
 wheels = [
@@ -3439,7 +3474,7 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/f2/2d/e38439b7f937e8bf91a9ff2b8d9713d0d8e64e980fc00e8d1945b8a5b74b/pyfftw-0.15.1.tar.gz", hash = "sha256:bbcde6d40d165e1cbaf12dde062ebfebe9e43394cac8c166e699ba2c9a4b0461", size = 192838, upload-time = "2025-10-22T19:58:56.683Z" }
 wheels = [
@@ -3586,13 +3621,13 @@ name = "python-ligo-lw"
 version = "1.8.4"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "ligo-segments", marker = "python_full_version < '3.11'" },
-    { name = "lscsoft-glue", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "python-dateutil", marker = "python_full_version < '3.11'" },
-    { name = "pyyaml", marker = "python_full_version < '3.11'" },
-    { name = "six", marker = "python_full_version < '3.11'" },
-    { name = "tqdm", marker = "python_full_version < '3.11'" },
+    { name = "ligo-segments" },
+    { name = "lscsoft-glue" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "python-dateutil" },
+    { name = "pyyaml" },
+    { name = "six" },
+    { name = "tqdm" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/13/b9/bc552b9b2cd0b0bddffae90aa7f1b95229d323dadb742a48576c7a67ab3f/python_ligo_lw-1.8.4.tar.gz", hash = "sha256:1ccaefeeda2a1cddca22b021ec88f550e24f87293a6f9753b3912e41cf0c95d1", size = 2428804, upload-time = "2025-06-30T14:37:58.437Z" }
 
@@ -3890,7 +3925,7 @@ name = "roman-numerals-py"
 version = "4.1.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "roman-numerals", marker = "python_full_version >= '3.11'" },
+    { name = "roman-numerals" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/cb/b5/de96fca640f4f656eb79bbee0e79aeec52e3e0e359f8a3e6a0d366378b64/roman_numerals_py-4.1.0.tar.gz", hash = "sha256:f5d7b2b4ca52dd855ef7ab8eb3590f428c0b1ea480736ce32b01fef2a5f8daf9", size = 4274, upload-time = "2025-12-17T18:25:41.153Z" }
 wheels = [
@@ -4008,10 +4043,10 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "joblib", marker = "python_full_version < '3.11'" },
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
-    { name = "threadpoolctl", marker = "python_full_version < '3.11'" },
+    { name = "joblib" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
+    { name = "scipy", version = "1.15.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "threadpoolctl" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/98/c2/a7855e41c9d285dfe86dc50b250978105dce513d6e459ea66a6aeb0e1e0c/scikit_learn-1.7.2.tar.gz", hash = "sha256:20e9e49ecd130598f1ca38a1d85090e1a600147b9c02fa6f15d69cb53d968fda", size = 7193136, upload-time = "2025-09-09T08:21:29.075Z" }
 wheels = [
@@ -4053,10 +4088,10 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "joblib", marker = "python_full_version >= '3.11'" },
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
-    { name = "threadpoolctl", marker = "python_full_version >= '3.11'" },
+    { name = "joblib" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
+    { name = "scipy", version = "1.16.3", source = { registry = "https://pypi.org/simple" } },
+    { name = "threadpoolctl" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/0e/d4/40988bf3b8e34feec1d0e6a051446b1f66225f8529b9309becaeef62b6c4/scikit_learn-1.8.0.tar.gz", hash = "sha256:9bccbb3b40e3de10351f8f5068e105d0f4083b1a65fa07b6634fbc401a6287fd", size = 7335585, upload-time = "2025-12-10T07:08:53.618Z" }
 wheels = [
@@ -4095,7 +4130,7 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/0f/37/6964b830433e654ec7485e45a00fc9a27cf868d622838f6b6d9c5ec0d532/scipy-1.15.3.tar.gz", hash = "sha256:eae3cf522bc7df64b42cad3925c876e1b0b6c35c1337c93e12c0f366f55b0eaf", size = 59419214, upload-time = "2025-05-08T16:13:05.955Z" }
 wheels = [
@@ -4157,7 +4192,7 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "numpy", version = "2.3.5", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/0a/ca/d8ace4f98322d01abcd52d381134344bf7b431eba7ed8b42bdea5a3c2ac9/scipy-1.16.3.tar.gz", hash = "sha256:01e87659402762f43bd2fee13370553a17ada367d42e7487800bf2916535aecb", size = 30597883, upload-time = "2025-10-28T17:38:54.068Z" }
 wheels = [
@@ -4212,7 +4247,7 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "optype", version = "0.9.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "optype", version = "0.9.3", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/0b/5f/35c43bd7d412add4adcd68475702571b2489b50c40b6564f808b2355e452/scipy_stubs-1.15.3.0.tar.gz", hash = "sha256:e8f76c9887461cf9424c1e2ad78ea5dac71dd4cbb383dc85f91adfe8f74d1e17", size = 275699, upload-time = "2025-05-08T16:58:35.139Z" }
 wheels = [
@@ -4230,7 +4265,7 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "optype", version = "0.15.0", source = { registry = "https://pypi.org/simple" }, extra = ["numpy"], marker = "python_full_version >= '3.11'" },
+    { name = "optype", version = "0.15.0", source = { registry = "https://pypi.org/simple" }, extra = ["numpy"] },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/08/91/1700d2a1a9f64f19bb019a547e510b99a6af1fef49641a0bce86bc85fb8e/scipy_stubs-1.16.3.3.tar.gz", hash = "sha256:af47578875d5557567225a16ec1b9b38a48c4c4377d92396413ebd65406c44ee", size = 361468, upload-time = "2025-12-08T13:45:38.37Z" }
 wheels = [
@@ -4400,23 +4435,23 @@ resolution-markers = [
     "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "alabaster", marker = "python_full_version < '3.11'" },
-    { name = "babel", marker = "python_full_version < '3.11'" },
-    { name = "colorama", marker = "python_full_version < '3.11' and sys_platform == 'win32'" },
-    { name = "docutils", marker = "python_full_version < '3.11'" },
-    { name = "imagesize", marker = "python_full_version < '3.11'" },
-    { name = "jinja2", marker = "python_full_version < '3.11'" },
-    { name = "packaging", marker = "python_full_version < '3.11'" },
-    { name = "pygments", marker = "python_full_version < '3.11'" },
-    { name = "requests", marker = "python_full_version < '3.11'" },
-    { name = "snowballstemmer", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-applehelp", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-devhelp", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-htmlhelp", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-jsmath", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-qthelp", marker = "python_full_version < '3.11'" },
-    { name = "sphinxcontrib-serializinghtml", marker = "python_full_version < '3.11'" },
-    { name = "tomli", marker = "python_full_version < '3.11'" },
+    { name = "alabaster" },
+    { name = "babel" },
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "docutils" },
+    { name = "imagesize" },
+    { name = "jinja2" },
+    { name = "packaging" },
+    { name = "pygments" },
+    { name = "requests" },
+    { name = "snowballstemmer" },
+    { name = "sphinxcontrib-applehelp" },
+    { name = "sphinxcontrib-devhelp" },
+    { name = "sphinxcontrib-htmlhelp" },
+    { name = "sphinxcontrib-jsmath" },
+    { name = "sphinxcontrib-qthelp" },
+    { name = "sphinxcontrib-serializinghtml" },
+    { name = "tomli" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/6f/6d/be0b61178fe2cdcb67e2a92fc9ebb488e3c51c4f74a36a7824c0adf23425/sphinx-8.1.3.tar.gz", hash = "sha256:43c1911eecb0d3e161ad78611bc905d1ad0e523e4ddc202a58a821773dc4c927", size = 8184611, upload-time = "2024-10-13T20:27:13.93Z" }
 wheels = [
@@ -4434,23 +4469,23 @@ resolution-markers = [
     "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
-    { name = "alabaster", marker = "python_full_version >= '3.11'" },
-    { name = "babel", marker = "python_full_version >= '3.11'" },
-    { name = "colorama", marker = "python_full_version >= '3.11' and sys_platform == 'win32'" },
-    { name = "docutils", marker = "python_full_version >= '3.11'" },
-    { name = "imagesize", marker = "python_full_version >= '3.11'" },
-    { name = "jinja2", marker = "python_full_version >= '3.11'" },
-    { name = "packaging", marker = "python_full_version >= '3.11'" },
-    { name = "pygments", marker = "python_full_version >= '3.11'" },
-    { name = "requests", marker = "python_full_version >= '3.11'" },
-    { name = "roman-numerals-py", marker = "python_full_version >= '3.11'" },
-    { name = "snowballstemmer", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-applehelp", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-devhelp", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-htmlhelp", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-jsmath", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-qthelp", marker = "python_full_version >= '3.11'" },
-    { name = "sphinxcontrib-serializinghtml", marker = "python_full_version >= '3.11'" },
+    { name = "alabaster" },
+    { name = "babel" },
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "docutils" },
+    { name = "imagesize" },
+    { name = "jinja2" },
+    { name = "packaging" },
+    { name = "pygments" },
+    { name = "requests" },
+    { name = "roman-numerals-py" },
+    { name = "snowballstemmer" },
+    { name = "sphinxcontrib-applehelp" },
+    { name = "sphinxcontrib-devhelp" },
+    { name = "sphinxcontrib-htmlhelp" },
+    { name = "sphinxcontrib-jsmath" },
+    { name = "sphinxcontrib-qthelp" },
+    { name = "sphinxcontrib-serializinghtml" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/38/ad/4360e50ed56cb483667b8e6dadf2d3fda62359593faabbe749a27c4eaca6/sphinx-8.2.3.tar.gz", hash = "sha256:398ad29dee7f63a75888314e9424d40f52ce5a6a87ae88e7071e80af296ec348", size = 8321876, upload-time = "2025-03-02T22:31:59.658Z" }
 wheels = [

From 84e3063ce797061bbd7c7a71ef6a59fecd46f7f1 Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Wed, 1 Jul 2026 16:09:46 +0200
Subject: [PATCH 08/15] WIP: Import logger and call setup_logger in dingo_pipe
 main (issue #152)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 dingo/pipe/main.py | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/dingo/pipe/main.py b/dingo/pipe/main.py
index 1c52cedcf..9bdb00bd1 100644
--- a/dingo/pipe/main.py
+++ b/dingo/pipe/main.py
@@ -16,12 +16,15 @@
     convert_string_to_dict,
     get_colored_string,
     get_command_line_arguments,
-    logger,
     parse_args,
     convert_prior_string_input,
     BilbyPipeError,
 )
 
+import logging
+from dingo.core.utils.logging_utils import setup_logger
+logger = logging.getLogger("dingo.pipe")
+
 from .dag_creator import generate_dag
 from .parser import create_parser
 from .utils import dict_to_string
@@ -618,6 +621,8 @@ def main():
     parser = create_parser(top_level=True)
     args, unknown_args = parse_args(get_command_line_arguments(), parser)
 
+    setup_logger(outdir=args.outdir, label=args.label)
+
     importance_sampling_updates, model_args = fill_in_arguments_from_model(args)
     inputs = MainInput(args, unknown_args, importance_sampling_updates)
     write_complete_config_file(parser, args, inputs)

From 5c77054d6f17063f95db6ffb5321ab87575d404a Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Thu, 2 Jul 2026 10:34:29 +0200
Subject: [PATCH 09/15] adding setup_logger to create log files

---
 dingo/gw/dataset/generate_dataset.py       | 4 +++-
 dingo/gw/dataset/generate_dataset_dag.py   | 3 ++-
 dingo/gw/dataset/utils.py                  | 7 ++++++-
 dingo/gw/noise/generate_dataset.py         | 3 ++-
 dingo/gw/training/train_pipeline.py        | 3 ++-
 dingo/gw/training/train_pipeline_condor.py | 3 ++-
 6 files changed, 17 insertions(+), 6 deletions(-)

diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index 873568c13..4e5886cad 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -24,7 +24,7 @@
     generate_waveforms_parallel,
 )
 from dingo.core.utils.misc import call_func_strict_output_dim
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 
 
 def generate_parameters_and_polarizations(
@@ -302,6 +302,8 @@ def _generate_dataset_main(
 
 def main() -> None:
     args = parse_args()
+    out_path = Path(args.out_file)
+    setup_logger(outdir=str(out_path.parent), label=out_path.stem)
     _generate_dataset_main(args.settings_file, args.out_file, args.num_processes)
 
 
diff --git a/dingo/gw/dataset/generate_dataset_dag.py b/dingo/gw/dataset/generate_dataset_dag.py
index c24c74b32..718149f72 100644
--- a/dingo/gw/dataset/generate_dataset_dag.py
+++ b/dingo/gw/dataset/generate_dataset_dag.py
@@ -5,7 +5,7 @@
 import yaml
 import copy
 
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 
 # Fixed file names
 svd_fn = "svd.hdf5"
@@ -266,6 +266,7 @@ def create_dag(args, settings):
 
 def main():
     args = parse_args()
+    setup_logger(outdir=args.temp_dir, label="generate_dataset_dag")
 
     # Load settings
     with open(args.settings_file, "r") as f:
diff --git a/dingo/gw/dataset/utils.py b/dingo/gw/dataset/utils.py
index aa5e84855..5308ef320 100644
--- a/dingo/gw/dataset/utils.py
+++ b/dingo/gw/dataset/utils.py
@@ -1,12 +1,13 @@
 import argparse
 import textwrap
 import copy
+from pathlib import Path
 import pandas as pd
 import numpy as np
 import yaml
 from typing import List
 
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 from dingo.gw.SVD import SVDBasis
 from dingo.gw.dataset.generate_dataset import train_svd_basis
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
@@ -88,6 +89,8 @@ def merge_datasets_cli():
         "--settings_file", type=str, help="YAML file containing new dataset settings."
     )
     args = parser.parse_args()
+    out_path = Path(args.out_file)
+    setup_logger(outdir=str(out_path.parent), label=out_path.stem)
 
     dataset_list = []
     for i in range(args.num_parts):
@@ -142,6 +145,8 @@ def build_svd_cli():
         "Remainder are used for validation.",
     )
     args = parser.parse_args()
+    out_path = Path(args.out_file)
+    setup_logger(outdir=str(out_path.parent), label=out_path.stem)
 
     dataset = WaveformDataset(file_name=args.dataset_file)
     if args.num_train is None:
diff --git a/dingo/gw/noise/generate_dataset.py b/dingo/gw/noise/generate_dataset.py
index 2a4c8b383..24f5c34a8 100644
--- a/dingo/gw/noise/generate_dataset.py
+++ b/dingo/gw/noise/generate_dataset.py
@@ -12,7 +12,7 @@
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.generate_dataset_dag import create_dag
 from dingo.gw.noise.utils import merge_datasets, get_time_segments
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 
 
 def parse_args():
@@ -59,6 +59,7 @@ def generate_dataset():
     Creates and saves an ASD dataset
     """
     args = parse_args()
+    setup_logger(outdir=args.data_dir, label="asd_dataset")
 
     # Load settings
     settings_file = (
diff --git a/dingo/gw/training/train_pipeline.py b/dingo/gw/training/train_pipeline.py
index 544f4b1ac..5668c389b 100644
--- a/dingo/gw/training/train_pipeline.py
+++ b/dingo/gw/training/train_pipeline.py
@@ -27,7 +27,7 @@
     build_train_and_test_loaders,
 )
 from dingo.core.utils.trainutils import EarlyStopping
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 from dingo.gw.dataset import WaveformDataset
 from dingo.core.posterior_models import BasePosteriorModel
 
@@ -439,6 +439,7 @@ def train_local():
     args = parse_args()
 
     os.makedirs(args.train_dir, exist_ok=True)
+    setup_logger(outdir=args.train_dir, label="dingo_train")
 
     if args.settings_file is not None:
         logger.info("Beginning new training run.")
diff --git a/dingo/gw/training/train_pipeline_condor.py b/dingo/gw/training/train_pipeline_condor.py
index 95f118add..06550f5a2 100644
--- a/dingo/gw/training/train_pipeline_condor.py
+++ b/dingo/gw/training/train_pipeline_condor.py
@@ -4,7 +4,7 @@
 import yaml
 import argparse
 
-from dingo.core.utils.logging_utils import logger
+from dingo.core.utils.logging_utils import logger, setup_logger
 from dingo.gw.training import (
     prepare_training_new,
     prepare_training_resume,
@@ -79,6 +79,7 @@ def train_condor():
         help="Optional command to execute after completion of training.",
     )
     args = parser.parse_args()
+    setup_logger(outdir=args.train_dir, label="dingo_train")
 
     if not args.start_submission:
 

From 08ad8004264f7a0bb861419da34c9ff3a8d50ce1 Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Thu, 2 Jul 2026 10:52:57 +0200
Subject: [PATCH 10/15] calling setup_logger to create log files

---
 dingo/pipe/plot.py     | 2 ++
 dingo/pipe/sampling.py | 5 ++++-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/dingo/pipe/plot.py b/dingo/pipe/plot.py
index dbd69de0c..772fa68f9 100644
--- a/dingo/pipe/plot.py
+++ b/dingo/pipe/plot.py
@@ -3,6 +3,7 @@
 
 import logging
 
+from dingo.core.utils.logging_utils import setup_logger
 from dingo.gw.result import Result
 
 logger = logging.getLogger("dingo.pipe")
@@ -48,6 +49,7 @@ def create_parser():
 
 def main():
     args, _ = parse_args(get_command_line_arguments(), create_parser())
+    setup_logger(outdir=args.outdir, label=args.label)
 
     logger.info(f"Generating plots for results file {args.result}")
 
diff --git a/dingo/pipe/sampling.py b/dingo/pipe/sampling.py
index 9138066bc..e038e7022 100644
--- a/dingo/pipe/sampling.py
+++ b/dingo/pipe/sampling.py
@@ -7,10 +7,12 @@
 from bilby_pipe.input import Input
 from bilby_pipe.utils import (
     parse_args,
-    logger,
     convert_string_to_dict,
     resolve_filename_with_transfer_fallback,
 )
+import logging
+from dingo.core.utils.logging_utils import setup_logger
+logger = logging.getLogger("dingo.pipe")
 
 from dingo.core.posterior_models.build_model import build_model_from_kwargs
 from dingo.gw.data.event_dataset import EventDataset
@@ -216,6 +218,7 @@ def create_sampling_parser():
 def main():
     """Data analysis main logic"""
     args, unknown_args = parse_args(sys.argv[1:], create_sampling_parser())
+    setup_logger(outdir=args.outdir, label=args.label)
     # log_version_information()
     analysis = SamplingInput(args, unknown_args)
     analysis.run_sampler()

From 974f15d32c562b856c7cfadcaeece2b7ad1255d1 Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Thu, 2 Jul 2026 10:54:39 +0200
Subject: [PATCH 11/15] Replace bilby_pipe logger with dingo.pipe logger and
 call setup_logger

---
 dingo/pipe/data_generation.py     | 6 ++++--
 dingo/pipe/importance_sampling.py | 7 ++++---
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/dingo/pipe/data_generation.py b/dingo/pipe/data_generation.py
index c8117027a..e465b3865 100644
--- a/dingo/pipe/data_generation.py
+++ b/dingo/pipe/data_generation.py
@@ -10,12 +10,12 @@
 from bilby.core.prior import PriorDict
 from bilby_pipe.utils import (
     parse_args,
-    logger,
     convert_string_to_dict,
     convert_prior_string_input,
     resolve_filename_with_transfer_fallback,
     BilbyPipeError,
 )
+import logging
 import lalsimulation as LS
 import numpy as np
 
@@ -24,8 +24,9 @@
 from dingo.gw.domains import UniformFrequencyDomain, build_domain_from_model_metadata
 from dingo.gw.injection import Injection
 from dingo.pipe.parser import create_parser
+from dingo.core.utils.logging_utils import setup_logger
 
-logger.name = "dingo_pipe"
+logger = logging.getLogger("dingo.pipe")
 
 
 class DataGenerationInput(BilbyDataGenerationInput):
@@ -534,6 +535,7 @@ def create_generation_parser():
 def main():
     """Data generation main logic"""
     args, unknown_args = parse_args(sys.argv[1:], create_generation_parser())
+    setup_logger(outdir=args.outdir, label=args.label)
     # log_version_information()
     data = DataGenerationInput(args, unknown_args)
     data.save_hdf5()
diff --git a/dingo/pipe/importance_sampling.py b/dingo/pipe/importance_sampling.py
index 541e8ebb2..bb195a647 100644
--- a/dingo/pipe/importance_sampling.py
+++ b/dingo/pipe/importance_sampling.py
@@ -8,12 +8,14 @@
 from bilby_pipe.input import Input
 from bilby_pipe.utils import (
     parse_args,
-    logger,
     convert_string_to_dict,
     convert_prior_string_input,
     resolve_filename_with_transfer_fallback,
     BilbyPipeError,
 )
+import logging
+from dingo.core.utils.logging_utils import setup_logger
+logger = logging.getLogger("dingo.pipe")
 
 from dingo.gw.data.event_dataset import EventDataset
 from dingo.gw.domains import MultibandedFrequencyDomain
@@ -21,8 +23,6 @@
 from dingo.pipe.parser import create_parser
 from dingo.gw.result import Result
 
-logger.name = "dingo_pipe"
-
 
 class ImportanceSamplingInput(Input):
     def __init__(self, args, unknown_args):
@@ -285,6 +285,7 @@ def create_sampling_parser():
 def main():
     """Data analysis main logic"""
     args, unknown_args = parse_args(sys.argv[1:], create_sampling_parser())
+    setup_logger(outdir=args.outdir, label=args.label)
     # log_version_information()
     analysis = ImportanceSamplingInput(args, unknown_args)
     analysis.run_sampler()

From 92651e472758b006875317a47a33c397a41cc052 Mon Sep 17 00:00:00 2001
From: cmfabbri <cecilia.fabbri@yahoo.it>
Date: Thu, 2 Jul 2026 12:05:15 +0200
Subject: [PATCH 12/15] add logger name to log file

---
 dingo/core/utils/logging_utils.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/dingo/core/utils/logging_utils.py b/dingo/core/utils/logging_utils.py
index 0de9d14e8..dd21cd426 100644
--- a/dingo/core/utils/logging_utils.py
+++ b/dingo/core/utils/logging_utils.py
@@ -77,6 +77,12 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
     if bilby_logger is not None:
         for handler in bilby_logger.handlers:
             if isinstance(handler, logging.FileHandler):
+                handler.setFormatter(
+                    logging.Formatter(
+                        "%(asctime)s %(name)s %(levelname)-8s: %(message)s",
+                        datefmt="%H:%M",
+                    )
+                )
                 if not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
                     logger.addHandler(handler)
 
@@ -90,7 +96,7 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
         file_handler = logging.FileHandler(log_file)
         file_handler.setFormatter(
             logging.Formatter(
-                "%(asctime)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
+                "%(asctime)s %(name)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
             )
         )
         file_handler.setLevel(level)

From 859bfee49928a023fc1fb299116a874a9dcf7ac8 Mon Sep 17 00:00:00 2001
From: Heinrich Campe <49278231+hcampe@users.noreply.github.com>
Date: Fri, 3 Jul 2026 16:21:25 +0200
Subject: [PATCH 13/15] stages 1,2 done; code not fully working yet, but the
 structure is there

---
 HYDRA_TODO.md                                 | 376 +++++++++++++-----
 configs/__init__.py                           |   0
 configs/append_training_stage.yaml            |  28 ++
 configs/domain/multibanded_frequency.yaml     |   8 +
 configs/domain/uniform_frequency.yaml         |   4 +
 configs/estimate_psds.yaml                    |  33 ++
 configs/evaluate_multibanded_domain.yaml      |  13 +
 configs/experiment/asd_fiducial.yaml          |  13 +
 configs/experiment/asd_full.yaml              |  13 +
 configs/experiment/asd_toy.yaml               |  13 +
 configs/experiment/generate_fmpe_dataset.yaml |  16 +
 configs/experiment/generate_gnpe_dataset.yaml |  16 +
 configs/experiment/generate_npe_dataset.yaml  |  16 +
 configs/experiment/generate_toy_dataset.yaml  |  14 +
 configs/experiment/importance_sampling.yaml   |  16 +
 configs/experiment/synthetic_asd.yaml         |  17 +
 configs/experiment/train_fmpe.yaml            |  65 +++
 configs/experiment/train_gnpe.yaml            |  72 ++++
 configs/experiment/train_gnpe_init.yaml       |  57 +++
 configs/experiment/train_npe.yaml             |  65 +++
 configs/experiment/train_toy.yaml             |  45 +++
 configs/extrinsic_prior/default.yaml          |   5 +
 configs/extrinsic_prior/fmpe.yaml             |   5 +
 configs/generate_asd_dataset.yaml             |  35 ++
 configs/generate_dataset.yaml                 |  15 +
 configs/generate_synthetic_asd_dataset.yaml   |  33 ++
 configs/hydra/default.yaml                    |   6 +
 configs/importance_weights.yaml               |  47 +++
 configs/intrinsic_prior/default.yaml          |  10 +
 configs/intrinsic_prior/fmpe.yaml             |  14 +
 configs/intrinsic_prior/precessing.yaml       |  14 +
 .../precessing_multibanded_test.yaml          |  14 +
 configs/merge_asd_datasets.yaml               |  35 ++
 configs/model/fmpe.yaml                       |  31 ++
 configs/model/gnpe.yaml                       |  21 +
 configs/model/gnpe_init.yaml                  |  21 +
 configs/model/npe.yaml                        |  21 +
 configs/model/toy_npe.yaml                    |  21 +
 configs/model/unconditional_npe.yaml          |  12 +
 configs/optimizer/adam.yaml                   |   2 +
 configs/scheduler/cosine.yaml                 |   2 +
 configs/train.yaml                            |  61 +++
 configs/unconditional_density_estimation.yaml |  29 ++
 configs/waveform_generator/phenom_d.yaml      |   4 +
 configs/waveform_generator/phenom_pv2.yaml    |   4 +
 configs/waveform_generator/phenom_xphm.yaml   |   4 +
 .../unconditional_density_estimation.py       |  44 +-
 dingo/core/posterior_models/base_model.py     |  21 +-
 dingo/core/result.py                          |  43 +-
 dingo/core/samplers.py                        |  36 +-
 dingo/core/utils/pt_to_hdf5.py                |  93 +++--
 dingo/core/utils/trainutils.py                |  46 ++-
 dingo/gw/SVD.py                               |  25 +-
 .../gw/dataset/evaluate_multibanded_domain.py |  89 ++---
 dingo/gw/dataset/generate_dataset.py          |   3 +-
 dingo/gw/dataset/utils.py                     |  41 +-
 dingo/gw/importance_sampling/diagnostics.py   |   5 +-
 .../importance_sampling/importance_weights.py | 104 ++---
 dingo/gw/inference/gw_samplers.py             |  13 +-
 dingo/gw/ls_cli.py                            | 116 +++---
 dingo/gw/noise/asd_dataset.py                 |   9 +-
 dingo/gw/noise/asd_estimation.py              |  85 ++--
 dingo/gw/noise/generate_dataset.py            | 115 ++----
 dingo/gw/noise/synthetic/asd_sampling.py      |  12 +-
 dingo/gw/noise/synthetic/generate_dataset.py  |  91 ++---
 dingo/gw/noise/utils.py                       |  72 ++--
 dingo/gw/result.py                            |  39 +-
 dingo/gw/training/train_builders.py           |  37 +-
 dingo/gw/training/train_pipeline.py           | 154 ++++---
 dingo/gw/training/utils.py                    |  50 +--
 dingo/gw/transforms/waveform_transforms.py    |   8 +-
 .../waveform_generator/waveform_generator.py  |   6 +-
 old_examples/__init__.py                      |   0
 old_examples/append_training_stage.yaml       |  28 ++
 old_examples/estimate_psds.yaml               |  29 ++
 old_examples/evaluate_multibanded_domain.yaml |  45 +++
 old_examples/generate_asd_dataset.yaml        |  31 ++
 .../generate_dataset.yaml                     |   0
 .../generate_synthetic_asd_dataset.yaml       |  31 ++
 old_examples/importance_weights.yaml          |  57 +++
 old_examples/merge_asd_datasets.yaml          |  31 ++
 old_examples/train.yaml                       |  84 ++++
 .../unconditional_density_estimation.yaml     |  42 ++
 pyproject.toml                                |   2 +-
 84 files changed, 2356 insertions(+), 747 deletions(-)
 create mode 100644 configs/__init__.py
 create mode 100644 configs/append_training_stage.yaml
 create mode 100644 configs/domain/multibanded_frequency.yaml
 create mode 100644 configs/domain/uniform_frequency.yaml
 create mode 100644 configs/estimate_psds.yaml
 create mode 100644 configs/evaluate_multibanded_domain.yaml
 create mode 100644 configs/experiment/asd_fiducial.yaml
 create mode 100644 configs/experiment/asd_full.yaml
 create mode 100644 configs/experiment/asd_toy.yaml
 create mode 100644 configs/experiment/generate_fmpe_dataset.yaml
 create mode 100644 configs/experiment/generate_gnpe_dataset.yaml
 create mode 100644 configs/experiment/generate_npe_dataset.yaml
 create mode 100644 configs/experiment/generate_toy_dataset.yaml
 create mode 100644 configs/experiment/importance_sampling.yaml
 create mode 100644 configs/experiment/synthetic_asd.yaml
 create mode 100644 configs/experiment/train_fmpe.yaml
 create mode 100644 configs/experiment/train_gnpe.yaml
 create mode 100644 configs/experiment/train_gnpe_init.yaml
 create mode 100644 configs/experiment/train_npe.yaml
 create mode 100644 configs/experiment/train_toy.yaml
 create mode 100644 configs/extrinsic_prior/default.yaml
 create mode 100644 configs/extrinsic_prior/fmpe.yaml
 create mode 100644 configs/generate_asd_dataset.yaml
 create mode 100644 configs/generate_dataset.yaml
 create mode 100644 configs/generate_synthetic_asd_dataset.yaml
 create mode 100644 configs/hydra/default.yaml
 create mode 100644 configs/importance_weights.yaml
 create mode 100644 configs/intrinsic_prior/default.yaml
 create mode 100644 configs/intrinsic_prior/fmpe.yaml
 create mode 100644 configs/intrinsic_prior/precessing.yaml
 create mode 100644 configs/intrinsic_prior/precessing_multibanded_test.yaml
 create mode 100644 configs/merge_asd_datasets.yaml
 create mode 100644 configs/model/fmpe.yaml
 create mode 100644 configs/model/gnpe.yaml
 create mode 100644 configs/model/gnpe_init.yaml
 create mode 100644 configs/model/npe.yaml
 create mode 100644 configs/model/toy_npe.yaml
 create mode 100644 configs/model/unconditional_npe.yaml
 create mode 100644 configs/optimizer/adam.yaml
 create mode 100644 configs/scheduler/cosine.yaml
 create mode 100644 configs/train.yaml
 create mode 100644 configs/unconditional_density_estimation.yaml
 create mode 100644 configs/waveform_generator/phenom_d.yaml
 create mode 100644 configs/waveform_generator/phenom_pv2.yaml
 create mode 100644 configs/waveform_generator/phenom_xphm.yaml
 create mode 100644 old_examples/__init__.py
 create mode 100644 old_examples/append_training_stage.yaml
 create mode 100644 old_examples/estimate_psds.yaml
 create mode 100644 old_examples/evaluate_multibanded_domain.yaml
 create mode 100644 old_examples/generate_asd_dataset.yaml
 rename {configs/examples => old_examples}/generate_dataset.yaml (100%)
 create mode 100644 old_examples/generate_synthetic_asd_dataset.yaml
 create mode 100644 old_examples/importance_weights.yaml
 create mode 100644 old_examples/merge_asd_datasets.yaml
 create mode 100644 old_examples/train.yaml
 create mode 100644 old_examples/unconditional_density_estimation.yaml

diff --git a/HYDRA_TODO.md b/HYDRA_TODO.md
index ecc322c7a..f4186cf51 100644
--- a/HYDRA_TODO.md
+++ b/HYDRA_TODO.md
@@ -40,6 +40,17 @@ one pass.
      defaults in Hydra configs.
    - Revisit saved metadata format and backward compatibility deliberately.
 
+## Stage 1 Scope
+
+Hydra Stage 1 covers scripts that consume substantial YAML/settings configs or
+launch run-like workflows where a run directory, saved config, and Hydra
+overrides are useful.
+
+Tiny inspection, conversion, compatibility, debug/demo, and maintenance scripts
+with only a few command-line arguments are intentionally allowed to stay as
+argparse scripts. They should still use logging for user-facing progress where
+we touch them, but they do not need Hydra configs.
+
 ## Stage 1 Definition Of Done
 
 For each migrated CLI:
@@ -54,6 +65,13 @@ For each migrated CLI:
       existing business logic.
 - [ ] Settings saved into HDF5/PT/result metadata remain loadable.
 - [ ] User-facing progress messages use `logging` instead of `print`.
+- [ ] Any `print(...)` statements reachable from the migrated CLI are replaced
+      by logging calls, including prints in helper modules called by that CLI.
+- [ ] Python warnings emitted by migrated CLIs are routed through logging, e.g.
+      with `logging.captureWarnings(True)`, so they appear in both stdout and
+      the Hydra log consistently.
+- [ ] Smoke tests compare captured stdout with the Hydra log file; they should
+      match exactly, confirming all progress output went through logging.
 - [ ] A small smoke test exists or is documented.
 
 Recommended shared pattern:
@@ -65,7 +83,7 @@ import hydra
 from omegaconf import DictConfig, OmegaConf
 
 log = logging.getLogger(__name__)
-
+logging.captureWarnings(True)
 
 @hydra.main(version_base="1.3", config_path="...", config_name="...")
 def main(cfg: DictConfig) -> None:
@@ -89,22 +107,107 @@ in the Hydra run directory together with `generate_dataset.log` and `.hydra/`
 metadata. The run root is controlled by the required `DINGO_RUNS` environment
 variable.
 
+When running local Stage 1 smoke tests, set `DINGO_RUNS` to the repo-local
+`my_runs/` directory so the generated run directories are easy to inspect:
+
+```bash
+DINGO_RUNS=/home/hcampe/dingo/my_runs
+```
+
+For migrated CLIs, capture stdout during smoke tests and compare it against the
+Hydra log file in the corresponding run directory. For example:
+
+```bash
+diff -u captured_stdout.txt my_runs/<run_dir>/<job_name>.log
+```
+
 If a user explicitly wants old current-working-directory behavior, use both:
 
 ```bash
 hydra.job.chdir=false hydra.run.dir=.
 ```
 
-Testing showed that `hydra.job.chdir=false` alone is not enough: application
-outputs land in the current working directory, but Hydra logs and `.hydra/`
-metadata still go to `hydra.run.dir`.
-
 Use `compression: null` rather than `compression: None`.
 
 Do not migrate `.ini` files for now. In particular, leave Asimov/plugin-style
 `.ini` configuration outside the Hydra migration TODO unless this is reopened
 explicitly later.
 
+## Known Follow-Ups
+
+- [ ] Decide on the long-term config package layout. For Stage 1,
+      `configs/` is made importable and included in setuptools package
+      discovery so console scripts can use ordinary relative
+      `@hydra.main(config_path=...)` values.
+      In Stage 2/3, decide whether to keep this top-level `configs` package,
+      move configs under `dingo/`, or introduce a shared config-search-path
+      helper/plugin.
+
+## Stage 1 Status
+
+Stage 1 is complete for the scoped set of settings/config-driven workflows.
+Remaining parser-based scripts are explicitly excluded below because they are
+small utilities, compatibility helpers, debug/demo mains, `.ini`-based pipe
+commands, or Condor/DAG submission paths deferred to Stage 3.
+
+## Stage 2 Status
+
+Stage 2 has started. The Stage 1 `configs/examples/` folder was renamed to
+`configs/old_examples/`, and migrated command configs now live directly under
+`configs/`.
+
+Current group structure:
+
+- `domain`
+- `waveform_generator`
+- `intrinsic_prior`
+- `extrinsic_prior`
+- `model`
+- `optimizer`
+- `scheduler`
+- `hydra`
+- `experiment`
+
+The group configs contain default settings, including several defaults that were
+previously implicit in examples or Python helper code. The `experiment` group
+contains overrides for documented/example workflows such as toy NPE, NPE, GNPE,
+GNPE initialization, FMPE, ASD variants, synthetic ASD generation, and
+importance sampling.
+
+The single `domain/uniform_frequency.yaml` entry is now the high-resolution
+uniform-frequency variant (`delta_f: 0.125`). Workflows that intentionally use
+the old coarser domain, such as the toy waveform dataset example, override
+`domain.delta_f` downstream.
+
+Posterior-model and embedding-model settings have been unified into a single
+`model` group, with one file per model family. The toy-sized normalizing-flow
+configuration from the documentation is named `model/toy_npe.yaml`; the
+production NPE settings are named `model/npe.yaml`; the unconditional density
+estimator used by density recovery/importance weighting is named
+`model/unconditional_npe.yaml`.
+
+Stage 2 audit notes:
+
+- The old examples and documentation are now represented by experiment configs
+  for toy NPE, production NPE, GNPE, GNPE initialization, FMPE, ASD variants,
+  synthetic ASD generation, and importance sampling.
+- The examples' explicit unconditional NDE settings are represented by
+  `model/unconditional_npe.yaml` and used by `importance_weights.yaml` and
+  `unconditional_density_estimation.yaml`.
+- Training-stage `early_stopping: null` is explicit in the base stage and the
+  documented experiment stages.
+- Condor resource blocks from production training examples are kept only as
+  settings data where they already existed in examples. Hydra-aware Condor
+  submission remains Stage 3 work.
+
+## Spotted Bugs
+
+- [ ] `dingo_build_svd`: the old `num_train is None` branch used
+      `len(WaveformDataset)`, i.e. the class object, which raises
+      `TypeError: object of type 'type' has no len()`. This was changed to
+      `len(dataset)`, the loaded instance length. Mention this deliberate bug
+      fix in the PR even though the CLI itself is no longer Hydra-migrated.
+
 ## Entrypoint Inventory
 
 Public console scripts are defined in `pyproject.toml`.
@@ -129,73 +232,112 @@ Public console scripts are defined in `pyproject.toml`.
   - Estimated Stage 1 difficulty: out of scope.
   - Estimated Stage 3 difficulty: high.
 
-- [ ] `dingo_evaluate_multibanded_domain`
+- [x] `dingo_evaluate_multibanded_domain`
   - File: `dingo/gw/dataset/evaluate_multibanded_domain.py`
-  - Stage 1: straightforward wrapper around current settings dict.
+  - Stage 1 implemented: `main` is wrapped with Hydra and uses
+    `configs/examples/evaluate_multibanded_domain.yaml`.
+  - This remains in scope because it has a structured domain, waveform
+    generator, prior, compression, and sample-count config.
   - Stage 3: replace `build_domain`, `build_prior_with_defaults`, and manual
     waveform-generator branching with Hydra targets.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: medium.
 
-- [ ] `dingo_build_svd`
+- [x] `dingo_build_svd`
   - File: `dingo/gw/dataset/utils.py`
-  - Stage 1: small CLI migration.
+  - Explicit Stage 1 exception: keep as a small argparse utility because it has
+    only a few CLI arguments and does not consume a full settings config.
+  - Still uses logging for user-facing progress.
+  - Keeps the deliberate bug fix in the old `num_train is None` path: use
+    `len(dataset)` rather than `len(WaveformDataset)`.
   - Stage 3: likely becomes part of a Hydra-instantiated compression/SVD
     workflow.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: medium.
 
-- [ ] `dingo_merge_datasets`
+- [x] `dingo_merge_datasets`
   - File: `dingo/gw/dataset/utils.py`
-  - Stage 1: small CLI migration.
+  - Explicit Stage 1 exception: keep as a small argparse utility. It has only a
+    few CLI arguments; the optional `settings_file` is a metadata override, not
+    a run-defining config.
+  - Uses logging for user-facing progress.
   - Stage 3: probably little to do.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: low.
 
 ### Noise / ASD Generation
 
-- [ ] `dingo_generate_asd_dataset`
+- [x] `dingo_generate_asd_dataset`
   - File: `dingo/gw/noise/generate_dataset.py`
-  - Stage 1: replace parser/settings-file handling with Hydra.
-  - Stage 1 should cover local generation only.
-  - Condor/DAG behavior should remain as-is or be deferred.
+  - Stage 1 implemented for local generation: `generate_dataset` is wrapped
+    directly with Hydra and uses `configs/examples/generate_asd_dataset.yaml`.
+  - Parser/settings-file handling was replaced by a shallow Hydra config that
+    preserves the old settings dict shape internally.
+  - Condor/DAG behavior is explicitly deferred to Stage 3 under Hydra.
+  - Uses `logging.captureWarnings(True)` and logs user-facing progress instead
+    of printing.
+  - Smoke test output is in
+    `my_runs/2026-07-01_14-43-30_generate_asd_dataset_dataset_settings.T=1.0,dataset_settings.detectors=[H1,L1],dataset_settings.f_s=16,dataset_settings.observing_run=O1,dataset_settings.time_psd=4`.
   - Estimated Stage 1 difficulty: medium.
   - Estimated Stage 3 difficulty: medium/high for Condor integration.
 
-- [ ] `dingo_estimate_psds`
+- [x] `dingo_estimate_psds`
   - File: `dingo/gw/noise/asd_estimation.py`
-  - Stage 1: straightforward wrapper around current dict settings.
+  - Stage 1 implemented: `download_and_estimate_cli` is wrapped directly with
+    Hydra and uses `configs/examples/estimate_psds.yaml`.
+  - The existing `download_and_estimate_psds(...)` dict-based implementation is
+    unchanged.
+  - Uses `logging.captureWarnings(True)`.
+  - Smoke test output is in
+    `my_runs/2026-07-01_14-55-15_estimate_psds_dataset_settings.T=1.0,dataset_settings.detectors=[H1,L1],dataset_settings.f_s=16,dataset_settings.observing_run=O1,dataset_settings.time_psd=4`.
   - Stage 3: domain construction and PSD-estimation settings could become more
     declarative.
   - Estimated Stage 1 difficulty: low/medium.
   - Estimated Stage 3 difficulty: medium.
 
-- [ ] `dingo_generate_synthetic_asd_dataset`
+- [x] `dingo_generate_synthetic_asd_dataset`
   - File: `dingo/gw/noise/synthetic/generate_dataset.py`
-  - Stage 1: straightforward config migration.
+  - Stage 1 implemented: `main` is wrapped with Hydra and uses
+    `configs/examples/generate_synthetic_asd_dataset.yaml`.
+  - The existing `generate_dataset(...)` implementation remains dict-based.
+  - Uses `logging.captureWarnings(True)`.
+  - Smoke test output is in
+    `my_runs/2026-07-01_14-44-49_generate_synthetic_asd_dataset_parameterization_settings.num_spectral_segments=2,parameterization_settings.num_spline_positions=3,sampling_settings=null`.
   - Stage 3: parameterization and sampling could become separate config groups.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: medium.
 
-- [ ] `dingo_merge_asd_datasets`
+- [x] `dingo_merge_asd_datasets`
   - File: `dingo/gw/noise/utils.py`
-  - Stage 1: small CLI migration.
+  - Stage 1 implemented: `merge_datasets_cli` is wrapped directly with Hydra
+    and uses `configs/examples/merge_asd_datasets.yaml`.
+  - Uses `logging.captureWarnings(True)` and logs user-facing progress instead
+    of printing.
+  - Smoke test output is in
+    `my_runs/2026-07-01_14-43-31_merge_asd_datasets_dataset_settings.T=1.0,dataset_settings.detectors=[H1,L1],dataset_settings.f_s=16,dataset_settings.observing_run=O1,dataset_settings.time_psd=4`.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: low.
 
 ### Training
 
-- [ ] `dingo_train`
+- [x] `dingo_train`
   - File: `dingo/gw/training/train_pipeline.py`
-  - Stage 1: replace parser with Hydra, while still converting to the old
-    `train_settings` and `local_settings` dicts.
-  - Watch out:
-    - New training versus checkpoint resume are currently mutually exclusive CLI
-      modes.
-    - `local` settings are popped out and saved separately as
-      `local_settings.yaml`.
-    - Train settings are saved into model metadata.
-    - Several progress messages still use `print`.
+  - Stage 1 implemented: `train_local` is wrapped directly with Hydra and uses
+    `configs/examples/train.yaml`.
+  - New training uses the Hydra config body as the old `train_settings` dict;
+    checkpoint resume uses `checkpoint=<path>`.
+  - `local` settings are still popped out and saved separately as
+    `local_settings.yaml`.
+  - Input dataset paths are resolved before training so `hydra.job.chdir=true`
+    does not break file loading; these resolved paths are saved in model
+    metadata.
+  - Training progress reachable from this CLI now uses logging instead of
+    printing, including `train_builders`, `BasePosteriorModel.train`, and
+    `trainutils`.
+  - Training omits `hydra.job.override_dirname` from its run directory template
+    because real training overrides are commonly too large for filesystem-safe
+    paths.
+  - Smoke test output is in `my_runs/2026-07-01_15-26-09_train`.
   - Estimated Stage 1 difficulty: medium/high.
   - Estimated Stage 3 difficulty: high.
 
@@ -210,83 +352,121 @@ Public console scripts are defined in `pyproject.toml`.
   - Estimated Stage 1 difficulty: out of scope.
   - Estimated Stage 3 difficulty: high.
 
-- [ ] `dingo_append_training_stage`
+- [x] `dingo_append_training_stage`
   - File: `dingo/gw/training/utils.py`
-  - Stage 1: small Hydra CLI.
-  - Watch out: modifies model metadata/training-stage settings.
+  - Stage 1 implemented: `append_stage` is wrapped directly with Hydra and
+    uses `configs/examples/append_training_stage.yaml`.
+  - The new stage is provided directly in the Hydra config under `stage`.
+  - Modifies model metadata/training-stage settings as before.
+  - Smoke test output is in
+    `my_runs/2026-07-01_15-26-55_append_training_stage_stage.batch_size=1,stage.epochs=2`.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: low/medium.
 
 ### Pipe / Inference
 
-The `dingo_pipe*` family is the largest migration area because it is based on
-`BilbyArgParser` / `configargparse` and has bilby_pipe-style conventions.
-
-- [ ] `dingo_pipe`
-  - File: `dingo/pipe/main.py`
-  - Uses parser from `dingo/pipe/parser.py`.
-  - Estimated Stage 1 difficulty: high.
-  - Estimated Stage 3 difficulty: very high.
-
-- [ ] `dingo_pipe_generation`
-  - File: `dingo/pipe/data_generation.py`
-  - Estimated Stage 1 difficulty: high.
-  - Estimated Stage 3 difficulty: high.
-
-- [ ] `dingo_pipe_sampling`
-  - File: `dingo/pipe/sampling.py`
-  - Estimated Stage 1 difficulty: high.
-  - Estimated Stage 3 difficulty: high.
-
-- [ ] `dingo_pipe_importance_sampling`
-  - File: `dingo/pipe/importance_sampling.py`
-  - Estimated Stage 1 difficulty: high.
-  - Estimated Stage 3 difficulty: high.
-
-- [ ] `dingo_pipe_plot`
-  - File: `dingo/pipe/plot.py`
-  - Estimated Stage 1 difficulty: medium/high.
-  - Estimated Stage 3 difficulty: medium.
-
-- [ ] `dingo_pipe_pp_test`
-  - File: `dingo/pipe/pp_test.py`
-  - Estimated Stage 1 difficulty: low/medium.
-  - Estimated Stage 3 difficulty: medium.
-
-- [ ] `dingo_result`
-  - File: `dingo/pipe/dingo_result.py`
-  - Small CLI, but connected to pipe output conventions.
-  - Estimated Stage 1 difficulty: low.
-  - Estimated Stage 3 difficulty: low.
+Explicitly out of scope for this Hydra migration for now. Keep the
+`dingo_pipe*` family, `dingo_result`, and their `.ini` / bilby_pipe-style
+configuration conventions as they are unless this decision is reopened later.
 
 ### Other Utilities
 
-- [ ] `dingo_pt_to_hdf5`
+- [x] `dingo_pt_to_hdf5`
   - File: `dingo/core/utils/pt_to_hdf5.py`
-  - Small argparse CLI.
+  - Explicit Stage 1 exception: keep as a small argparse conversion utility
+    because it has only `in_file`, `out_file`, and `model_version_number`.
+  - Uses `logging.captureWarnings(True)` and stream-only logging instead of
+    printing, without Hydra and without saving a log file.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: low.
 
-- [ ] `dingo_ls`
+- [x] `dingo_ls`
   - File: `dingo/gw/ls_cli.py`
-  - Inspection tool rather than a config-driven workflow.
-  - Could stay argparse for a while, but migrate for consistency if Stage 1 aims
-    to eliminate all public parsers.
+  - Explicit Stage 1 exception: keep this as a small argparse inspection script
+    because it has only one positional argument and is not a config-driven
+    workflow.
+  - Uses `logging.captureWarnings(True)` and stream-only logging instead of
+    printing, without Hydra and without saving a log file. The stream formatter
+    matches the standard Hydra log format.
+  - `Result.print_summary()` was updated to use logging because it is reachable
+    from this CLI.
+  - Smoke test output was captured in
+    `/tmp/dingo_ls_stdout_standard_format.out`.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: low.
 
-- [ ] Importance-weight generation
+- [x] Importance-weight generation
   - File: `dingo/gw/importance_sampling/importance_weights.py`
-  - Not currently listed in `project.scripts`, but it is a config-driven CLI.
+  - Stage 1 implemented: `main` is wrapped with Hydra and uses
+    `configs/examples/importance_weights.yaml`.
+  - The old `--settings` / YAML-loading path was removed; the Hydra config body
+    is converted to a plain settings dict internally.
+  - Input parameter-sample and calibration-envelope paths are resolved before
+    use so `hydra.job.chdir=true` does not break file loading.
+  - User-facing progress reachable from this CLI now uses logging instead of
+    printing, including sampler progress, result summaries, and diagnostics.
+  - Config-composition smoke was captured in
+    `/tmp/dingo_importance_weights_cfg.out`. A full run still needs a
+    physically meaningful GW result and likelihood setup.
   - Estimated Stage 1 difficulty: low/medium.
   - Estimated Stage 3 difficulty: medium.
 
-- [ ] Unconditional density estimation
+- [x] Unconditional density estimation
   - File: `dingo/core/density/unconditional_density_estimation.py`
-  - Not currently listed in `project.scripts`, but has an argparse settings CLI.
+  - Stage 1 implemented: the stale `--settings` parser was replaced by a Hydra
+    entry point using `configs/examples/unconditional_density_estimation.yaml`.
+  - The Hydra config takes `result_file`, `train_dir`, and the old estimator
+    settings; internally the estimator still receives a plain settings dict.
+  - Smoke test output is in
+    `my_runs/unconditional_density_estimation_stage1_smoke`.
   - Estimated Stage 1 difficulty: low.
   - Estimated Stage 3 difficulty: medium.
 
+### Explicitly Not Ported To Hydra
+
+These are intentionally excluded from Stage 1. Mention them in the PR so it is
+clear this is a scoped migration rather than an incomplete sweep.
+
+- Condor / DAG submission:
+  - `dingo_generate_dataset_dag`
+    (`dingo/gw/dataset/generate_dataset_dag.py`)
+  - `dingo_train_condor`
+    (`dingo/gw/training/train_pipeline_condor.py`)
+  - Noise-generation DAG helper
+    (`dingo/gw/noise/generate_dataset_dag.py`)
+  - `dingo/core/utils/condor_utils.py`
+  - Rationale: Condor integration belongs in Stage 3, ideally as a
+    Hydra-aware submission path rather than shallow argparse replacement.
+- Pipe / `.ini` workflows:
+  - `dingo_pipe`
+  - `dingo_pipe_generation`
+  - `dingo_pipe_sampling`
+  - `dingo_pipe_importance_sampling`
+  - `dingo_pipe_plot`
+  - `dingo_pipe_pp_test`
+  - `dingo_result`
+  - Rationale: keep bilby_pipe-style `.ini` conventions for now.
+- Small public utilities kept as argparse:
+  - `dingo_ls`
+  - `dingo_pt_to_hdf5`
+  - `dingo_build_svd`
+  - `dingo_merge_datasets`
+  - Rationale: only a few command-line arguments and no full run-defining
+    settings config. These may still use logging for progress output.
+- Compatibility scripts:
+  - `compatibility/*.py`
+  - Rationale: one-off saved-artifact migration helpers.
+- Miscellaneous maintenance scripts:
+  - `misc_scripts/*.py`
+  - Rationale: ad hoc plotting, monitoring, evaluation, and maintenance tools.
+    Some take a settings file, but they are not public run-launching workflows.
+- Debug/demo module mains:
+  - `dingo/gw/likelihood.py`
+  - `dingo/gw/waveform_generator/waveform_generator.py`
+  - `dingo/core/nn/enets.py`
+  - `dingo/core/nn/nsf.py`
+  - Rationale: local debug/demo entry points, not user-facing config workflows.
+
 ## Stage 2 Config Group Candidates
 
 - `domain`
@@ -297,13 +477,10 @@ The `dingo_pipe*` family is the largest migration area because it is based on
 - `dataset`
 - `asd_dataset`
 - `noise`
-- `model/posterior`
-- `model/embedding`
+- `model`
 - `training/stage`
 - `training/local`
 - `condor` (Stage 3 only)
-- `pipe/event`
-- `pipe/sampler`
 - `importance_sampling`
 - `defaults`
 - `defaults/prior`
@@ -338,14 +515,33 @@ The `dingo_pipe*` family is the largest migration area because it is based on
     as long as old code handles them.
   - Candidate areas:
     - waveform generator optional kwargs such as `f_start`, `mode_list`,
-      `spin_conversion_phase`, and interface-specific options
+      `spin_conversion_phase`, `new_interface`, `lmax_nyquist`, and
+      interface-specific options
     - dataset compression/SVD/whitening options
     - training local/runtime/checkpoint options
     - optimizer and scheduler defaults
-    - transform defaults in the training data pipeline
+    - transform defaults in the training data pipeline, including
+      `zero_noise: false`, optional `random_strain_cropping`, derived
+      `context_parameters`, and derived `standardization`
     - pipe/inference defaults currently supplied by parser defaults
     - ASD/noise generation defaults such as frequency range, channels, and
       Condor settings
+    - dataloader defaults such as fixed split seed, `pin_memory`, worker seed
+      initialization, and `persistent_workers`
+    - early stopping defaults (`patience`, `verbose`, `delta`, `metric`)
+    - runtime limit defaults (`max_time_per_run`, `max_epochs_per_run`,
+      `max_epochs_total`, `epoch_start`)
+
+- [ ] `dingo/core/density/nde_settings.py`
+  - Move or retire `get_default_nde_settings_3d(...)` and
+    `DEFAULT_NDE_SETTINGS_2D`.
+  - These currently define additional unconditional NDE architectures:
+    - 3D: 5 flow steps, hidden dimension 128, batch size 4096, 10 epochs,
+      Adam learning rate 0.005, cosine `T_max: 10`.
+    - 2D GNPE proxy recovery: 5 flow steps, hidden dimension 64, batch size
+      4096, 10 epochs, Adam learning rate 0.001, cosine `T_max: 10`.
+  - Decide whether these become separate `model/*` and training config groups,
+    or whether they are superseded by `model/unconditional_npe.yaml`.
 
 - [ ] `dingo/gw/domains/build_domain.py`
   - Replace `type`-based dispatch with Hydra instantiation.
@@ -400,6 +596,8 @@ The `dingo_pipe*` family is the largest migration area because it is based on
     - `ProxyRecoveryDefault`
     - `PhaseRecoveryDefault`
     - `MultibandingDefault`
+  - These are intentionally deferred while `.ini`/pipe workflows remain outside
+    the Stage 1/2 Hydra migration.
 
 - [ ] `dingo/core/density/nde_settings.py`
   - Move `get_default_nde_settings_3d(...)` and `DEFAULT_NDE_SETTINGS_2D` into
diff --git a/configs/__init__.py b/configs/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/configs/append_training_stage.yaml b/configs/append_training_stage.yaml
new file mode 100644
index 000000000..88bcd1c58
--- /dev/null
+++ b/configs/append_training_stage.yaml
@@ -0,0 +1,28 @@
+defaults:
+  - hydra: default
+  - optimizer: adam
+  - scheduler: cosine
+  - _self_
+
+checkpoint: ???
+out_file: updated_model.pt
+replace: null
+
+stage:
+  optimizer: ${optimizer}
+  scheduler: ${scheduler}
+  epochs: 20
+  asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+  freeze_rb_layer: false
+  batch_size: 64
+  early_stopping: null
+
+hydra:
+  job:
+    name: append_training_stage
+    config:
+      override_dirname:
+        exclude_keys:
+          - checkpoint
+          - out_file
+          - stage.asd_dataset_path
diff --git a/configs/domain/multibanded_frequency.yaml b/configs/domain/multibanded_frequency.yaml
new file mode 100644
index 000000000..9b9289d68
--- /dev/null
+++ b/configs/domain/multibanded_frequency.yaml
@@ -0,0 +1,8 @@
+type: MultibandedFrequencyDomain
+nodes: [20.0, 26.0, 34.0, 46.0, 62.0, 78.0, 1038.0]
+delta_f_initial: 0.0625
+base_domain:
+  type: UniformFrequencyDomain
+  f_min: 20.0
+  f_max: 1037.9375
+  delta_f: 0.0625
diff --git a/configs/domain/uniform_frequency.yaml b/configs/domain/uniform_frequency.yaml
new file mode 100644
index 000000000..cb44035cd
--- /dev/null
+++ b/configs/domain/uniform_frequency.yaml
@@ -0,0 +1,4 @@
+type: UniformFrequencyDomain
+f_min: 20.0
+f_max: 1024.0
+delta_f: 0.125
diff --git a/configs/estimate_psds.yaml b/configs/estimate_psds.yaml
new file mode 100644
index 000000000..36e58ca2f
--- /dev/null
+++ b/configs/estimate_psds.yaml
@@ -0,0 +1,33 @@
+defaults:
+  - hydra: default
+  - _self_
+
+data_dir: asd_dataset
+time_segments_file: ???
+verbose: false
+
+dataset_settings:
+  f_min: 0
+  f_max: 2048
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  channels: null
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  job:
+    name: estimate_psds
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
diff --git a/configs/evaluate_multibanded_domain.yaml b/configs/evaluate_multibanded_domain.yaml
new file mode 100644
index 000000000..f372aa01f
--- /dev/null
+++ b/configs/evaluate_multibanded_domain.yaml
@@ -0,0 +1,13 @@
+defaults:
+  - hydra: default
+  - domain: multibanded_frequency
+  - waveform_generator: phenom_xphm
+  - intrinsic_prior: precessing_multibanded_test
+  - _self_
+
+num_samples: 1
+compression: null
+
+hydra:
+  job:
+    name: evaluate_multibanded_domain
diff --git a/configs/experiment/asd_fiducial.yaml b/configs/experiment/asd_fiducial.yaml
new file mode 100644
index 000000000..d69d57528
--- /dev/null
+++ b/configs/experiment/asd_fiducial.yaml
@@ -0,0 +1,13 @@
+# @package _global_
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  detectors: [H1, L1]
+  observing_run: O1
diff --git a/configs/experiment/asd_full.yaml b/configs/experiment/asd_full.yaml
new file mode 100644
index 000000000..ca2a2704b
--- /dev/null
+++ b/configs/experiment/asd_full.yaml
@@ -0,0 +1,13 @@
+# @package _global_
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 0
+  detectors: [H1, L1]
+  observing_run: O1
diff --git a/configs/experiment/asd_toy.yaml b/configs/experiment/asd_toy.yaml
new file mode 100644
index 000000000..b1e86708e
--- /dev/null
+++ b/configs/experiment/asd_toy.yaml
@@ -0,0 +1,13 @@
+# @package _global_
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 4.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  detectors: [H1, L1]
+  observing_run: O1
diff --git a/configs/experiment/generate_fmpe_dataset.yaml b/configs/experiment/generate_fmpe_dataset.yaml
new file mode 100644
index 000000000..a62c125fe
--- /dev/null
+++ b/configs/experiment/generate_fmpe_dataset.yaml
@@ -0,0 +1,16 @@
+# @package _global_
+
+defaults:
+  - override /domain: uniform_frequency
+  - override /waveform_generator: phenom_pv2
+  - override /intrinsic_prior: fmpe
+  - _self_
+
+num_samples: 5000000
+compression:
+  svd:
+    size: 200
+    num_training_samples: 50000
+    num_validation_samples: 10000
+  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_gnpe_dataset.yaml b/configs/experiment/generate_gnpe_dataset.yaml
new file mode 100644
index 000000000..13cd02db9
--- /dev/null
+++ b/configs/experiment/generate_gnpe_dataset.yaml
@@ -0,0 +1,16 @@
+# @package _global_
+
+defaults:
+  - override /domain: uniform_frequency
+  - override /waveform_generator: phenom_xphm
+  - override /intrinsic_prior: precessing
+  - _self_
+
+num_samples: 5000000
+compression:
+  svd:
+    size: 200
+    num_training_samples: 50000
+    num_validation_samples: 10000
+  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_npe_dataset.yaml b/configs/experiment/generate_npe_dataset.yaml
new file mode 100644
index 000000000..ebc141c8e
--- /dev/null
+++ b/configs/experiment/generate_npe_dataset.yaml
@@ -0,0 +1,16 @@
+# @package _global_
+
+defaults:
+  - override /domain: uniform_frequency
+  - override /waveform_generator: phenom_xphm
+  - override /intrinsic_prior: default
+  - _self_
+
+num_samples: 5000000
+compression:
+  svd:
+    size: 200
+    num_training_samples: 50000
+    num_validation_samples: 10000
+  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_toy_dataset.yaml b/configs/experiment/generate_toy_dataset.yaml
new file mode 100644
index 000000000..386dac175
--- /dev/null
+++ b/configs/experiment/generate_toy_dataset.yaml
@@ -0,0 +1,14 @@
+# @package _global_
+
+defaults:
+  - override /domain: uniform_frequency
+  - override /waveform_generator: phenom_d
+  - override /intrinsic_prior: default
+  - _self_
+
+domain:
+  delta_f: 0.25
+num_samples: 10000
+compression: null
+num_processes: 1
+out_file: waveform_dataset.hdf5
diff --git a/configs/experiment/importance_sampling.yaml b/configs/experiment/importance_sampling.yaml
new file mode 100644
index 000000000..0ab8eb7c2
--- /dev/null
+++ b/configs/experiment/importance_sampling.yaml
@@ -0,0 +1,16 @@
+# @package _global_
+
+num_samples: 1000
+time_marginalization:
+  n_fft: 5
+slice_plots:
+  num_slice_plots: 10
+  params_slice2d:
+    - [phase, geocent_time]
+    - [phase, tilt_1]
+calibration_marginalization:
+  num_calibration_curves: 100
+  num_calibration_nodes: 10
+  calibration_envelope:
+    H1: /path/to/H1-calibration-envelope
+    L1: /path/to/L1-calibration-envelope
diff --git a/configs/experiment/synthetic_asd.yaml b/configs/experiment/synthetic_asd.yaml
new file mode 100644
index 000000000..91782b5bc
--- /dev/null
+++ b/configs/experiment/synthetic_asd.yaml
@@ -0,0 +1,17 @@
+# @package _global_
+
+parameterization_settings:
+  num_spline_positions: 30
+  num_spectral_segments: 400
+  sigma: 0.14
+  delta_f: -1
+  smoothen: true
+sampling_settings:
+  bandwidth_spectral: 0.5
+  bandwidth_spline: 0.25
+  split_frequencies:
+    - 30
+    - 100
+  rescaling_asd_paths:
+    H1: /path/to/rescaling_asd_H1.hdf5
+    L1: /path/to/rescaling_asd_L1.hdf5
diff --git a/configs/experiment/train_fmpe.yaml b/configs/experiment/train_fmpe.yaml
new file mode 100644
index 000000000..84eaca337
--- /dev/null
+++ b/configs/experiment/train_fmpe.yaml
@@ -0,0 +1,65 @@
+# @package _global_
+
+defaults:
+  - override /extrinsic_prior: fmpe
+  - override /model: fmpe
+  - _self_
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update:
+    f_min: 20.0
+    f_max: 512.0
+  svd_size_update: 150
+  detectors: [H1, L1]
+  ref_time: 1126259462.391
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - a_1
+    - a_2
+    - tilt_1
+    - tilt_2
+    - phi_12
+    - phi_jl
+    - theta_jn
+    - luminosity_distance
+    - geocent_time
+    - ra
+    - dec
+    - psi
+
+training:
+  evaluate: true
+  stage_0:
+    epochs: 400
+    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    freeze_rb_layer: true
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      lr: 0.0005
+    scheduler:
+      T_max: 400
+  stage_1:
+    epochs: 100
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: true
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      type: adam
+      lr: 1.0e-5
+    scheduler:
+      type: cosine
+      T_max: 100
+
+local:
+  device: cuda
+  num_workers: 31
+  runtime_limits:
+    max_time_per_run: 360000
+    max_epochs_per_run: 500
+  checkpoint_epochs: 200
+  leave_waveforms_on_disk: true
diff --git a/configs/experiment/train_gnpe.yaml b/configs/experiment/train_gnpe.yaml
new file mode 100644
index 000000000..d3342ae27
--- /dev/null
+++ b/configs/experiment/train_gnpe.yaml
@@ -0,0 +1,72 @@
+# @package _global_
+
+defaults:
+  - override /model: gnpe
+  - _self_
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update:
+    f_min: 20.0
+    f_max: 1024.0
+  svd_size_update: 200
+  detectors: [H1, L1]
+  ref_time: 1126259462.391
+  gnpe_time_shifts:
+    kernel: bilby.core.prior.Uniform(minimum=-0.001, maximum=0.001)
+    exact_equiv: true
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - a_1
+    - a_2
+    - tilt_1
+    - tilt_2
+    - phi_12
+    - phi_jl
+    - theta_jn
+    - luminosity_distance
+    - geocent_time
+    - ra
+    - dec
+    - psi
+
+training:
+  stage_0:
+    epochs: 300
+    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    freeze_rb_layer: true
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      lr: 5.0e-5
+    scheduler:
+      T_max: 300
+  stage_1:
+    epochs: 150
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: false
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      type: adam
+      lr: 1.0e-5
+    scheduler:
+      type: cosine
+      T_max: 150
+
+local:
+  device: cuda
+  num_workers: 32
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 500
+  checkpoint_epochs: 10
+  leave_waveforms_on_disk: true
+  condor:
+    num_cpus: 16
+    memory_cpus: 128000
+    num_gpus: 1
+    memory_gpus: 8000
+    request_disk: 50GB
diff --git a/configs/experiment/train_gnpe_init.yaml b/configs/experiment/train_gnpe_init.yaml
new file mode 100644
index 000000000..f47dd2cba
--- /dev/null
+++ b/configs/experiment/train_gnpe_init.yaml
@@ -0,0 +1,57 @@
+# @package _global_
+
+defaults:
+  - override /model: gnpe_init
+  - _self_
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update:
+    f_min: 20.0
+    f_max: 1024.0
+  svd_size_update: 200
+  detectors: [H1, L1]
+  ref_time: 1126259462.391
+  inference_parameters:
+    - H1_time
+    - L1_time
+
+training:
+  stage_0:
+    epochs: 150
+    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    freeze_rb_layer: true
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      lr: 5.0e-5
+    scheduler:
+      T_max: 150
+  stage_1:
+    epochs: 75
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: false
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      type: adam
+      lr: 1.0e-5
+    scheduler:
+      type: cosine
+      T_max: 75
+
+local:
+  device: cpu
+  num_workers: 6
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 500
+  checkpoint_epochs: 10
+  leave_waveforms_on_disk: true
+  condor:
+    num_cpus: 16
+    memory_cpus: 128000
+    num_gpus: 1
+    memory_gpus: 8000
+    request_disk: 50GB
diff --git a/configs/experiment/train_npe.yaml b/configs/experiment/train_npe.yaml
new file mode 100644
index 000000000..0ecd913ec
--- /dev/null
+++ b/configs/experiment/train_npe.yaml
@@ -0,0 +1,65 @@
+# @package _global_
+
+defaults:
+  - override /model: npe
+  - _self_
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update:
+    f_min: 20.0
+    f_max: 1024.0
+  svd_size_update: 200
+  detectors: [H1, L1]
+  ref_time: 1126259462.391
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - chi_1
+    - chi_2
+    - theta_jn
+    - dec
+    - ra
+    - geocent_time
+    - luminosity_distance
+    - psi
+
+training:
+  stage_0:
+    epochs: 300
+    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    freeze_rb_layer: true
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      lr: 5.0e-5
+    scheduler:
+      T_max: 300
+  stage_1:
+    epochs: 150
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: false
+    batch_size: 4096
+    early_stopping: null
+    optimizer:
+      type: adam
+      lr: 1.0e-5
+    scheduler:
+      type: cosine
+      T_max: 150
+
+local:
+  device: cuda
+  num_workers: 32
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 500
+  checkpoint_epochs: 50
+  leave_waveforms_on_disk: true
+  condor:
+    num_cpus: 16
+    memory_cpus: 128000
+    num_gpus: 1
+    memory_gpus: 8000
+    request_disk: 50GB
diff --git a/configs/experiment/train_toy.yaml b/configs/experiment/train_toy.yaml
new file mode 100644
index 000000000..687fad524
--- /dev/null
+++ b/configs/experiment/train_toy.yaml
@@ -0,0 +1,45 @@
+# @package _global_
+
+defaults:
+  - override /model: toy_npe
+  - _self_
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update: null
+  svd_size_update: null
+  detectors: [H1, L1]
+  ref_time: 1126259462.391
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - chi_1
+    - chi_2
+    - theta_jn
+    - dec
+    - ra
+    - geocent_time
+    - luminosity_distance
+    - psi
+    - phase
+
+training:
+  stage_0:
+    epochs: 20
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: true
+    batch_size: 64
+    early_stopping: null
+    optimizer:
+      lr: 0.0001
+    scheduler:
+      T_max: 20
+
+local:
+  device: cpu
+  num_workers: 6
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 30
+  checkpoint_epochs: 15
diff --git a/configs/extrinsic_prior/default.yaml b/configs/extrinsic_prior/default.yaml
new file mode 100644
index 000000000..702057e6d
--- /dev/null
+++ b/configs/extrinsic_prior/default.yaml
@@ -0,0 +1,5 @@
+dec: default
+ra: default
+geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
+psi: default
+luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/extrinsic_prior/fmpe.yaml b/configs/extrinsic_prior/fmpe.yaml
new file mode 100644
index 000000000..4d79f9f7e
--- /dev/null
+++ b/configs/extrinsic_prior/fmpe.yaml
@@ -0,0 +1,5 @@
+dec: default
+ra: default
+geocent_time: bilby.core.prior.Uniform(minimum=-0.03, maximum=0.03, name='geocent_time')
+psi: default
+luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/generate_asd_dataset.yaml b/configs/generate_asd_dataset.yaml
new file mode 100644
index 000000000..aba442e62
--- /dev/null
+++ b/configs/generate_asd_dataset.yaml
@@ -0,0 +1,35 @@
+defaults:
+  - hydra: default
+  - _self_
+
+data_dir: asd_dataset
+time_segments_file: null
+out_name: null
+verbose: false
+
+dataset_settings:
+  f_min: 0
+  f_max: 2048
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  channels: null
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  job:
+    name: generate_asd_dataset
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
+          - out_name
diff --git a/configs/generate_dataset.yaml b/configs/generate_dataset.yaml
new file mode 100644
index 000000000..05838a47b
--- /dev/null
+++ b/configs/generate_dataset.yaml
@@ -0,0 +1,15 @@
+defaults:
+  - hydra: default
+  - domain: uniform_frequency
+  - waveform_generator: phenom_d
+  - intrinsic_prior: default
+  - _self_
+
+num_samples: 10000
+compression: null
+num_processes: 1
+out_file: waveform_dataset.hdf5
+
+hydra:
+  job:
+    name: generate_dataset
diff --git a/configs/generate_synthetic_asd_dataset.yaml b/configs/generate_synthetic_asd_dataset.yaml
new file mode 100644
index 000000000..71070af80
--- /dev/null
+++ b/configs/generate_synthetic_asd_dataset.yaml
@@ -0,0 +1,33 @@
+defaults:
+  - hydra: default
+  - _self_
+
+asd_dataset: ???
+num_samples: 500
+num_processes: 1
+out_file: synthetic_asd_dataset.hdf5
+verbose: false
+
+parameterization_settings:
+  num_spline_positions: 30
+  num_spectral_segments: 400
+  sigma: 0.14
+  delta_f: -1
+  smoothen: true
+
+sampling_settings:
+  bandwidth_spectral: 0.5
+  bandwidth_spline: 0.25
+  split_frequencies:
+    - 30
+    - 100
+  rescaling_asd_paths: null
+
+hydra:
+  job:
+    name: generate_synthetic_asd_dataset
+    config:
+      override_dirname:
+        exclude_keys:
+          - asd_dataset
+          - out_file
diff --git a/configs/hydra/default.yaml b/configs/hydra/default.yaml
new file mode 100644
index 000000000..7b9f91d97
--- /dev/null
+++ b/configs/hydra/default.yaml
@@ -0,0 +1,6 @@
+# @package _global_
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    chdir: true
diff --git a/configs/importance_weights.yaml b/configs/importance_weights.yaml
new file mode 100644
index 000000000..7f924ac7a
--- /dev/null
+++ b/configs/importance_weights.yaml
@@ -0,0 +1,47 @@
+defaults:
+  - hydra: default
+  - model: unconditional_npe
+  - optimizer: adam
+  - scheduler: cosine
+  - _self_
+
+outdir: .
+parameter_samples: ???
+num_samples: 1000
+event_dataset: null
+time_marginalization:
+  n_fft: 5
+phase_marginalization: null
+synthetic_phase: null
+calibration_marginalization: null
+num_processes: 1
+
+slice_plots:
+  num_slice_plots: 10
+  params_slice2d:
+    - [phase, geocent_time]
+    - [phase, tilt_1]
+
+nde:
+  path: null
+  data:
+    parameter_samples: null
+    parameters: null
+  model: ${model}
+  training:
+    optimizer: ${optimizer}
+    scheduler: ${scheduler}
+    device: cpu
+    num_workers: 0
+    train_fraction: 0.9
+    batch_size: 512
+    epochs: 20
+
+hydra:
+  job:
+    name: importance_weights
+    config:
+      override_dirname:
+        exclude_keys:
+          - parameter_samples
+          - nde.path
diff --git a/configs/intrinsic_prior/default.yaml b/configs/intrinsic_prior/default.yaml
new file mode 100644
index 000000000..609f3d6e7
--- /dev/null
+++ b/configs/intrinsic_prior/default.yaml
@@ -0,0 +1,10 @@
+mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+phase: default
+chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
+chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
+theta_jn: default
+luminosity_distance: 100.0
+geocent_time: 0.0
diff --git a/configs/intrinsic_prior/fmpe.yaml b/configs/intrinsic_prior/fmpe.yaml
new file mode 100644
index 000000000..30d74ebb4
--- /dev/null
+++ b/configs/intrinsic_prior/fmpe.yaml
@@ -0,0 +1,14 @@
+mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
+mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
+chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=20.0, maximum=120.0, name='chirp_mass')
+mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+phase: default
+a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
+a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
+tilt_1: default
+tilt_2: default
+phi_12: default
+phi_jl: default
+theta_jn: default
+luminosity_distance: 100.0
+geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing.yaml b/configs/intrinsic_prior/precessing.yaml
new file mode 100644
index 000000000..5400d688d
--- /dev/null
+++ b/configs/intrinsic_prior/precessing.yaml
@@ -0,0 +1,14 @@
+mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
+mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
+chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=150.0, name='chirp_mass')
+mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+phase: default
+a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
+a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
+tilt_1: default
+tilt_2: default
+phi_12: default
+phi_jl: default
+theta_jn: default
+luminosity_distance: 100.0
+geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing_multibanded_test.yaml b/configs/intrinsic_prior/precessing_multibanded_test.yaml
new file mode 100644
index 000000000..7c3f469c1
--- /dev/null
+++ b/configs/intrinsic_prior/precessing_multibanded_test.yaml
@@ -0,0 +1,14 @@
+mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+phase: default
+a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
+a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
+tilt_1: default
+tilt_2: default
+phi_12: default
+phi_jl: default
+theta_jn: default
+luminosity_distance: 100.0
+geocent_time: 0.0
diff --git a/configs/merge_asd_datasets.yaml b/configs/merge_asd_datasets.yaml
new file mode 100644
index 000000000..1dbafebb1
--- /dev/null
+++ b/configs/merge_asd_datasets.yaml
@@ -0,0 +1,35 @@
+defaults:
+  - hydra: default
+  - _self_
+
+data_dir: asd_dataset
+time_segments_file: null
+num_parts: -1
+out_name: null
+
+dataset_settings:
+  f_min: 0
+  f_max: 2048
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  channels: null
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  job:
+    name: merge_asd_datasets
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
+          - out_name
diff --git a/configs/model/fmpe.yaml b/configs/model/fmpe.yaml
new file mode 100644
index 000000000..94241980e
--- /dev/null
+++ b/configs/model/fmpe.yaml
@@ -0,0 +1,31 @@
+posterior_model_type: flow_matching
+posterior_kwargs:
+  activation: gelu
+  batch_norm: true
+  context_with_glu: false
+  dropout: 0.0
+  hidden_dims: [4096, 4096, 4096, 2048, 2048, 2048, 1024, 1024, 1024, 1024, 1024, 1024, 512, 512, 512, 512, 512, 512, 512, 512, 256, 256, 256, 256, 256, 256, 256, 256, 256, 256, 128, 128, 128, 128, 128, 64, 64, 64, 32, 32, 32, 16, 16, 16]
+  sigma_min: 0.001
+  theta_embedding_kwargs:
+    embedding_net:
+      activation: gelu
+      hidden_dims: [16, 32, 64, 128, 256]
+      output_dim: 256
+      type: DenseResidualNet
+    encoding:
+      encode_all: false
+      frequencies: 0
+  theta_with_glu: true
+  time_prior_exponent: 1
+  type: DenseResidualNet
+embedding_kwargs:
+  activation: gelu
+  batch_norm: true
+  dropout: 0.0
+  hidden_dims:
+    - 2048
+  output_dim: 2048
+  svd:
+    num_training_samples: 50000
+    num_validation_samples: 10000
+    size: 150
diff --git a/configs/model/gnpe.yaml b/configs/model/gnpe.yaml
new file mode 100644
index 000000000..17bd48972
--- /dev/null
+++ b/configs/model/gnpe.yaml
@@ -0,0 +1,21 @@
+posterior_model_type: normalizing_flow
+posterior_kwargs:
+  num_flow_steps: 30
+  base_transform_kwargs:
+    hidden_dim: 1024
+    num_transform_blocks: 5
+    activation: elu
+    dropout_probability: 0.0
+    batch_norm: true
+    num_bins: 8
+    base_transform_type: rq-coupling
+embedding_kwargs:
+  output_dim: 128
+  hidden_dims: [1024, 1024, 1024, 1024, 1024, 1024, 512, 512, 512, 512, 512, 512, 256, 256, 256, 256, 256, 256, 128, 128, 128, 128, 128, 128]
+  activation: elu
+  dropout: 0.0
+  batch_norm: true
+  svd:
+    num_training_samples: 50000
+    num_validation_samples: 10000
+    size: 200
diff --git a/configs/model/gnpe_init.yaml b/configs/model/gnpe_init.yaml
new file mode 100644
index 000000000..ee39f8ec4
--- /dev/null
+++ b/configs/model/gnpe_init.yaml
@@ -0,0 +1,21 @@
+posterior_model_type: normalizing_flow
+posterior_kwargs:
+  num_flow_steps: 15
+  base_transform_kwargs:
+    hidden_dim: 512
+    num_transform_blocks: 5
+    activation: elu
+    dropout_probability: 0.0
+    batch_norm: true
+    num_bins: 8
+    base_transform_type: rq-coupling
+embedding_kwargs:
+  output_dim: 128
+  hidden_dims: [512, 512, 512, 512, 512, 512, 256, 256, 256, 256, 256, 256, 128, 128, 128, 128, 128, 128]
+  activation: elu
+  dropout: 0.0
+  batch_norm: true
+  svd:
+    num_training_samples: 50000
+    num_validation_samples: 10000
+    size: 200
diff --git a/configs/model/npe.yaml b/configs/model/npe.yaml
new file mode 100644
index 000000000..17bd48972
--- /dev/null
+++ b/configs/model/npe.yaml
@@ -0,0 +1,21 @@
+posterior_model_type: normalizing_flow
+posterior_kwargs:
+  num_flow_steps: 30
+  base_transform_kwargs:
+    hidden_dim: 1024
+    num_transform_blocks: 5
+    activation: elu
+    dropout_probability: 0.0
+    batch_norm: true
+    num_bins: 8
+    base_transform_type: rq-coupling
+embedding_kwargs:
+  output_dim: 128
+  hidden_dims: [1024, 1024, 1024, 1024, 1024, 1024, 512, 512, 512, 512, 512, 512, 256, 256, 256, 256, 256, 256, 128, 128, 128, 128, 128, 128]
+  activation: elu
+  dropout: 0.0
+  batch_norm: true
+  svd:
+    num_training_samples: 50000
+    num_validation_samples: 10000
+    size: 200
diff --git a/configs/model/toy_npe.yaml b/configs/model/toy_npe.yaml
new file mode 100644
index 000000000..295545298
--- /dev/null
+++ b/configs/model/toy_npe.yaml
@@ -0,0 +1,21 @@
+posterior_model_type: normalizing_flow
+posterior_kwargs:
+  num_flow_steps: 5
+  base_transform_kwargs:
+    hidden_dim: 64
+    num_transform_blocks: 5
+    activation: elu
+    dropout_probability: 0.0
+    batch_norm: true
+    num_bins: 8
+    base_transform_type: rq-coupling
+embedding_kwargs:
+  output_dim: 128
+  hidden_dims: [1024, 512, 256, 128]
+  activation: elu
+  dropout: 0.0
+  batch_norm: true
+  svd:
+    num_training_samples: 1000
+    num_validation_samples: 100
+    size: 50
diff --git a/configs/model/unconditional_npe.yaml b/configs/model/unconditional_npe.yaml
new file mode 100644
index 000000000..41c500b77
--- /dev/null
+++ b/configs/model/unconditional_npe.yaml
@@ -0,0 +1,12 @@
+posterior_model_type: normalizing_flow
+embedding_kwargs: null
+posterior_kwargs:
+  num_flow_steps: 10
+  base_transform_kwargs:
+    hidden_dim: 128
+    num_transform_blocks: 2
+    activation: elu
+    dropout_probability: 0.1
+    batch_norm: true
+    num_bins: 8
+    base_transform_type: rq-coupling
diff --git a/configs/optimizer/adam.yaml b/configs/optimizer/adam.yaml
new file mode 100644
index 000000000..30ba48b22
--- /dev/null
+++ b/configs/optimizer/adam.yaml
@@ -0,0 +1,2 @@
+type: adam
+lr: 0.0001
diff --git a/configs/scheduler/cosine.yaml b/configs/scheduler/cosine.yaml
new file mode 100644
index 000000000..c45a236f1
--- /dev/null
+++ b/configs/scheduler/cosine.yaml
@@ -0,0 +1,2 @@
+type: cosine
+T_max: 20
diff --git a/configs/train.yaml b/configs/train.yaml
new file mode 100644
index 000000000..631369b33
--- /dev/null
+++ b/configs/train.yaml
@@ -0,0 +1,61 @@
+defaults:
+  - hydra: default
+  - extrinsic_prior: default
+  - model: toy_npe
+  - optimizer: adam
+  - scheduler: cosine
+  - _self_
+
+checkpoint: null
+train_dir: train
+exit_command: ""
+
+data:
+  extrinsic_prior: ${extrinsic_prior}
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  domain_update: null
+  svd_size_update: null
+  detectors:
+    - H1
+    - L1
+  ref_time: 1126259462.391
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - chi_1
+    - chi_2
+    - theta_jn
+    - dec
+    - ra
+    - geocent_time
+    - luminosity_distance
+    - psi
+    - phase
+
+training:
+  evaluate: false
+  stage_0:
+    optimizer: ${optimizer}
+    scheduler: ${scheduler}
+    epochs: 20
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: true
+    batch_size: 64
+    early_stopping: null
+
+local:
+  device: cpu
+  num_workers: 6
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 30
+  checkpoint_epochs: 15
+  leave_waveforms_on_disk: true
+  local_cache_path: null
+  wandb: false
+  test_only: false
+
+hydra:
+  job:
+    name: train
diff --git a/configs/unconditional_density_estimation.yaml b/configs/unconditional_density_estimation.yaml
new file mode 100644
index 000000000..8129c0f5f
--- /dev/null
+++ b/configs/unconditional_density_estimation.yaml
@@ -0,0 +1,29 @@
+defaults:
+  - hydra: default
+  - model: unconditional_npe
+  - optimizer: adam
+  - scheduler: cosine
+  - _self_
+
+result_file: ???
+train_dir: nde
+
+data:
+  parameters: null
+
+training:
+  optimizer: ${optimizer}
+  scheduler: ${scheduler}
+  device: cpu
+  num_workers: 0
+  train_fraction: 0.9
+  batch_size: 512
+  epochs: 20
+
+hydra:
+  job:
+    name: unconditional_density_estimation
+    config:
+      override_dirname:
+        exclude_keys:
+          - result_file
diff --git a/configs/waveform_generator/phenom_d.yaml b/configs/waveform_generator/phenom_d.yaml
new file mode 100644
index 000000000..4ec9d64b3
--- /dev/null
+++ b/configs/waveform_generator/phenom_d.yaml
@@ -0,0 +1,4 @@
+approximant: IMRPhenomD
+f_ref: 20.0
+f_start: null
+spin_conversion_phase: null
diff --git a/configs/waveform_generator/phenom_pv2.yaml b/configs/waveform_generator/phenom_pv2.yaml
new file mode 100644
index 000000000..f4390b2fb
--- /dev/null
+++ b/configs/waveform_generator/phenom_pv2.yaml
@@ -0,0 +1,4 @@
+approximant: IMRPhenomPv2
+f_ref: 20.0
+f_start: null
+spin_conversion_phase: 0.0
diff --git a/configs/waveform_generator/phenom_xphm.yaml b/configs/waveform_generator/phenom_xphm.yaml
new file mode 100644
index 000000000..8c1a14c17
--- /dev/null
+++ b/configs/waveform_generator/phenom_xphm.yaml
@@ -0,0 +1,4 @@
+approximant: IMRPhenomXPHM
+f_ref: 20.0
+f_start: null
+spin_conversion_phase: 0.0
diff --git a/dingo/core/density/unconditional_density_estimation.py b/dingo/core/density/unconditional_density_estimation.py
index 53cd08858..48d056cae 100644
--- a/dingo/core/density/unconditional_density_estimation.py
+++ b/dingo/core/density/unconditional_density_estimation.py
@@ -1,14 +1,20 @@
 import copy
+import logging
+import os
 
+import hydra
 import torch
+import numpy as np
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from dingo.core.utils import build_train_and_test_loaders
 from dingo.core.utils.trainutils import RuntimeLimits
-import numpy as np
-import argparse
-
 from dingo.core.posterior_models import NormalizingFlowPosteriorModel
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
 
 class SampleDataset(torch.utils.data.Dataset):
     """
@@ -110,14 +116,24 @@ def train_unconditional_density_estimator(
     return model
 
 
-def parse_args():
-    parser = argparse.ArgumentParser(
-        description="Unconditional density estimation for dingo samples."
-    )
-    parser.add_argument(
-        "--settings",
-        type=str,
-        required=True,
-        help="Path to settings file.",
-    )
-    return parser.parse_args()
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="unconditional_density_estimation",
+)
+def main(cfg: DictConfig):
+    from dingo.core.result import Result
+
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    result_file = to_absolute_path(settings.pop("result_file"))
+    train_dir = settings.pop("train_dir")
+    os.makedirs(train_dir, exist_ok=True)
+
+    log.info(f"Loading result from {result_file}.")
+    result = Result(file_name=result_file)
+    log.info(f"Training unconditional density estimator in {train_dir}.")
+    train_unconditional_density_estimator(result, settings, train_dir)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/dingo/core/posterior_models/base_model.py b/dingo/core/posterior_models/base_model.py
index 68508d199..16e347f00 100755
--- a/dingo/core/posterior_models/base_model.py
+++ b/dingo/core/posterior_models/base_model.py
@@ -4,6 +4,7 @@
 """
 
 from abc import abstractmethod, ABC
+import logging
 import os
 from os.path import join
 import h5py
@@ -23,6 +24,8 @@
 
 from dingo.core.utils.trainutils import EarlyStopping
 
+log = logging.getLogger(__name__)
+
 
 class BasePosteriorModel(ABC):
     """
@@ -198,7 +201,7 @@ def network_to_device(self, device):
         #     raise NotImplementedError('This needs testing!')
         #     # dim = 0 [512, ...] -> [256, ...], [256, ...] on 2 GPUs
         #     self.network = torch.nn.DataParallel(self.network)
-        print(f"Putting posterior model to device {self.device}.")
+        log.info(f"Putting posterior model to device {self.device}.")
         self.network.to(self.device)
 
     def initialize_optimizer_and_scheduler(self):
@@ -394,7 +397,7 @@ def train(
 
         if test_only:
             test_loss = test_epoch(self, test_loader)
-            print(f"test loss: {test_loss:.3f}")
+            log.info(f"test loss: {test_loss:.3f}")
 
         else:
             while not runtime_limits.limits_exceeded(self.epoch):
@@ -403,24 +406,24 @@ def train(
                 # Training
                 lr = utils.get_lr(self.optimizer)
                 with threadpool_limits(limits=1, user_api="blas"):
-                    print(f"\nStart training epoch {self.epoch} with lr {lr}")
+                    log.info(f"\nStart training epoch {self.epoch} with lr {lr}")
                     time_start = time.time()
                     train_loss = train_epoch(self, train_loader)
                     train_time = time.time() - time_start
 
-                    print(
+                    log.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(train_time, 60)
                         )
                     )
 
                     # Testing
-                    print(f"Start testing epoch {self.epoch}")
+                    log.info(f"Start testing epoch {self.epoch}")
                     time_start = time.time()
                     test_loss = test_epoch(self, test_loader)
                     test_time = time.time() - time_start
 
-                    print(
+                    log.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(time.time() - time_start, 60)
                         )
@@ -449,7 +452,7 @@ def train(
                             }
                         )
                     except ImportError:
-                        print("wandb not installed. Skipping logging to wandb.")
+                        log.info("wandb not installed. Skipping logging to wandb.")
 
                 if early_stopping is not None:
                     # Whether to use train or test loss
@@ -464,9 +467,9 @@ def train(
                             join(train_dir, "best_model.pt"), save_training_info=False
                         )
                     if early_stopping.early_stop:
-                        print("Early stopping")
+                        log.info("Early stopping")
                         break
-                print(f"Finished training epoch {self.epoch}.\n")
+                log.info(f"Finished training epoch {self.epoch}.\n")
 
 
 def train_epoch(pm, dataloader):
diff --git a/dingo/core/result.py b/dingo/core/result.py
index 67017271a..8d7ad2a1e 100644
--- a/dingo/core/result.py
+++ b/dingo/core/result.py
@@ -1,4 +1,5 @@
 import copy
+import logging
 import math
 import tempfile
 import time
@@ -29,6 +30,8 @@
     "log_noise_evidence",
 ]
 
+log = logging.getLogger(__name__)
+
 
 def _clip_weights(weights: np.ndarray, num_clip: int) -> np.ndarray:
     """Replace the ``num_clip`` largest weights with their mean, then re-normalize.
@@ -167,23 +170,23 @@ def reset_event(self, event_dataset):
         ):
             # This is really just for notification. Actions are only taken if the
             # event metadata differ.
-            print("\nNew event data differ from existing.")
+            log.info("\nNew event data differ from existing.")
         self.context = context
 
         if self.event_metadata is not None and self.event_metadata != event_metadata:
-            print("Changes")
-            print("=======")
+            log.info("Changes")
+            log.info("=======")
             old_minus_new = dict(freeze(self.event_metadata) - freeze(event_metadata))
-            print("Old event metadata:")
+            log.info("Old event metadata:")
             for k in sorted(old_minus_new):
-                print(f"  {k}:  {self.event_metadata[k]}")
+                log.info(f"  {k}:  {self.event_metadata[k]}")
 
             new_minus_old = dict(freeze(event_metadata) - freeze(self.event_metadata))
-            print("New event metadata:")
+            log.info("New event metadata:")
             if self.importance_sampling_metadata.get("updates") is None:
                 self.importance_sampling_metadata["updates"] = {}
             for k in sorted(new_minus_old):
-                print(f"  {k}:  {event_metadata[k]}")
+                log.info(f"  {k}:  {event_metadata[k]}")
                 self.importance_sampling_metadata["updates"][k] = event_metadata[k]
 
             self._rebuild_domain(verbose=True)
@@ -312,12 +315,12 @@ def importance_sample(self, num_processes: int = 1, **likelihood_kwargs):
         valid_samples = np.isfinite(log_prior + delta_log_prob_target)
         theta = theta.iloc[valid_samples]
 
-        print(f"Calculating {len(theta)} likelihoods.")
+        log.info(f"Calculating {len(theta)} likelihoods.")
         t0 = time.time()
         log_likelihood = self.likelihood.log_likelihood_multi(
             theta, num_processes=num_processes
         )
-        print(f"Done. This took {time.time() - t0:.2f} seconds.")
+        log.info(f"Done. This took {time.time() - t0:.2f} seconds.")
 
         self.log_noise_evidence = self.likelihood.log_Zn
         self.samples["log_prior"] = log_prior
@@ -494,7 +497,7 @@ def rejection_sample(
         rng = np.random.default_rng(random_state)
         weights = self.samples["weights"].to_numpy(dtype=float)
 
-        print(
+        log.info(
             f"Rejection sampling: {self.num_samples} samples, "
             f"ESS = {self.effective_sample_size:.0f} "
             f"(efficiency = {100 * self.sample_efficiency:.1f}%)"
@@ -526,7 +529,7 @@ def rejection_sample(
         unweighted = unweighted.drop(
             columns=["weights", "log_prob", "delta_log_prob_target"], errors="ignore"
         )
-        print(f"Produced {len(unweighted)} unweighted samples.")
+        log.info(f"Produced {len(unweighted)} unweighted samples.")
         return unweighted
 
     def parameter_subset(self, parameters):
@@ -619,12 +622,12 @@ def print_summary(self):
         Display the number of samples, and (if importance sampling is complete) the log
         evidence and number of effective samples.
         """
-        print("Number of samples:", len(self.samples))
+        log.info(f"Number of samples: {len(self.samples)}")
         if self.log_evidence is not None:
-            print(
+            log.info(
                 f"Log(evidence): {self.log_evidence:.3f} +- {self.log_evidence_std:.3f}"
             )
-            print(
+            log.info(
                 f"Effective samples {self.n_eff:.1f}: "
                 f"(Sample efficiency = {100 * self.sample_efficiency:.2f}%)"
             )
@@ -815,7 +818,7 @@ def plot_log_probs(self, filename="log_probs.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            print("Results not importance sampled. Cannot produce log_prob plot.")
+            log.info("Results not importance sampled. Cannot produce log_prob plot.")
 
     def plot_weights(self, filename="weights.png"):
         """Make a scatter plot of samples weights vs log proposal."""
@@ -844,7 +847,7 @@ def plot_weights(self, filename="weights.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            print("Results not importance sampled. Cannot plot weights.")
+            log.info("Results not importance sampled. Cannot plot weights.")
 
     def get_all_injection_credible_levels(
         self, keys: list[str] = None, weighted: bool = False
@@ -1020,7 +1023,7 @@ def make_pp_plot(
 
     pvalues = []
     latex_labels = get_latex_labels(results[0].prior)
-    print("Key: KS-test p-value")
+    log.info("Key: KS-test p-value")
     for ii, key in enumerate(credible_levels):
         pp = np.array(
             [
@@ -1030,7 +1033,7 @@ def make_pp_plot(
         )
         pvalue = scipy.stats.kstest(credible_levels[key], "uniform").pvalue
         pvalues.append(pvalue)
-        print("{}: {}".format(key, pvalue))
+        log.info("{}: {}".format(key, pvalue))
         label = "{} ({:2.3f})".format(latex_labels[key], pvalue)
         plt.plot(x_values, pp, lines[ii], label=label, **kwargs)
 
@@ -1040,7 +1043,7 @@ def make_pp_plot(
         pvalues=pvalues,
         names=list(credible_levels.keys()),
     )
-    print("Combined p-value: {}".format(pvals.combined_pvalue))
+    log.info("Combined p-value: {}".format(pvals.combined_pvalue))
 
     if title:
         ax.set_title(
@@ -1090,5 +1093,3 @@ def freeze(d):
     elif isinstance(d, list):
         return tuple(freeze(value) for value in d)
     return d
-
-
diff --git a/dingo/core/samplers.py b/dingo/core/samplers.py
index efc4fc1fa..03e25b22f 100644
--- a/dingo/core/samplers.py
+++ b/dingo/core/samplers.py
@@ -1,9 +1,9 @@
 import copy
+import logging
 import math
 import time
 from pathlib import Path
 from typing import Optional, Union
-import sys
 
 import numpy as np
 import pandas as pd
@@ -22,6 +22,8 @@
 # Sampler classes are motivated by the approach of Bilby.
 #
 
+log = logging.getLogger(__name__)
+
 
 class Sampler(object):
     """
@@ -157,7 +159,7 @@ def _run_sampler(
             x = [x]
         else:
             if context is not None:
-                print("Unconditional model. Ignoring context.")
+                log.info("Unconditional model. Ignoring context.")
             x = []
 
         # For a normalizing flow, we get the log_prob for "free" when sampling,
@@ -205,7 +207,7 @@ def run_sampler(
         """
         self.samples = None
 
-        print(f"Running sampler to generate {num_samples} samples.")
+        log.info(f"Running sampler to generate {num_samples} samples.")
         t0 = time.time()
         if not self.unconditional_model:
             if self.context is None:
@@ -229,8 +231,7 @@ def run_sampler(
         # correction for t_ref) and represent as DataFrame.
         self._post_process(samples)
         self.samples = pd.DataFrame(samples)
-        print(f"Done. This took {time.time() - t0:.1f} s.")
-        sys.stdout.flush()
+        log.info(f"Done. This took {time.time() - t0:.1f} s.")
 
     def log_prob(self, samples: pd.DataFrame | dict) -> np.ndarray:
         """
@@ -430,7 +431,7 @@ def _run_sampler(
             init_samples = self.init_sampler._run_sampler(num_samples, context)
         else:
             if self.num_iterations == 1:
-                print(
+                log.info(
                     f"Warning: Removing initial outliers, but only carrying out "
                     f"{self.num_iterations} GNPE iteration. This risks biasing "
                     f"results."
@@ -515,16 +516,21 @@ def _run_sampler(
                 p: x["extrinsic_parameters"][p] for p in self.gnpe_proxy_parameters
             }
 
-            print(
+            log.info(
                 f"it {i}.\tmin pvalue: {self.iteration_tracker.pvalue_min:.3f}"
-                f"\tproxy mean: ",
-                *[f"{torch.mean(v).item():.5f}" for v in proxies.values()],
-                "\tproxy std:",
-                *[f"{torch.std(v).item():.5f}" for v in proxies.values()],
-                "\ttimes:",
-                time_sample_start - start_time,
-                time_sample_end - time_sample_start,
-                time.time() - time_sample_end,
+                f"\tproxy mean: "
+                + " ".join(f"{torch.mean(v).item():.5f}" for v in proxies.values())
+                + "\tproxy std: "
+                + " ".join(f"{torch.std(v).item():.5f}" for v in proxies.values())
+                + "\ttimes: "
+                + " ".join(
+                    str(t)
+                    for t in (
+                        time_sample_start - start_time,
+                        time_sample_end - time_sample_start,
+                        time.time() - time_sample_end,
+                    )
+                )
             )
 
         #
diff --git a/dingo/core/utils/pt_to_hdf5.py b/dingo/core/utils/pt_to_hdf5.py
index 06e6aa96c..195ecfe1f 100644
--- a/dingo/core/utils/pt_to_hdf5.py
+++ b/dingo/core/utils/pt_to_hdf5.py
@@ -1,38 +1,71 @@
 import os
+import argparse
+import logging
+import sys
+
 import torch
-import numpy as np
 import h5py
 import json
-import argparse
+
+
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
 
 def parse_args():
     parser = argparse.ArgumentParser(
-        description="Convert the weights of a trained Dingo model from a PyTorch pickle .pt file to HDF5,"
-        " for distribution in the LVK's CVMFS.",
-        epilog="Training history (optimizer_state_dict) is discarded.")
-    parser.add_argument("-i", "--in_file", type=str, required=True,
-            help='Input model ".pt" weights file')
-    parser.add_argument("-o", "--out_file", type=str, required=True,
-            help='Output model ".hdf5" weights file')
-    parser.add_argument("-n", "--model_version_number", type=int, required=True,
-            help="Model version number (integer). "
-            "Will be included in the output filename and metadata.")
+        description=(
+            "Convert the weights of a trained Dingo model from a PyTorch pickle "
+            ".pt file to HDF5, for distribution in the LVK's CVMFS."
+        ),
+        epilog="Training history (optimizer_state_dict) is discarded.",
+    )
+    parser.add_argument(
+        "-i",
+        "--in_file",
+        type=str,
+        required=True,
+        help='Input model ".pt" weights file',
+    )
+    parser.add_argument(
+        "-o", "--out_file", type=str, required=True, help='Output model ".hdf5" file'
+    )
+    parser.add_argument(
+        "-n",
+        "--model_version_number",
+        type=int,
+        required=True,
+        help=(
+            "Model version number (integer). Will be included in the output "
+            "filename and metadata."
+        ),
+    )
     return parser.parse_args()
 
 
+def configure_logging():
+    logging.basicConfig(
+        level=logging.INFO,
+        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
+        stream=sys.stdout,
+        force=True,
+    )
+
+
 def main():
+    configure_logging()
     args = parse_args()
-    if os.path.splitext(args.in_file)[-1] != '.pt':
-        raise ValueError('Expected a .pt input file')
-    if os.path.splitext(args.out_file)[-1] != '.hdf5':
-        raise ValueError('Expected a .hdf5 output file')
+
+    if os.path.splitext(args.in_file)[-1] != ".pt":
+        raise ValueError("Expected a .pt input file")
+    if os.path.splitext(args.out_file)[-1] != ".hdf5":
+        raise ValueError("Expected a .hdf5 output file")
 
     # Build output filename with the version number for this network
     # This is required for use on CVMFS
     root, ext = os.path.splitext(args.out_file)
-    out_file_name = f'{root}_v{args.model_version_number}{ext}'
-    print('Output will be written to', out_file_name)
+    out_file_name = f"{root}_v{args.model_version_number}{ext}"
+    log.info(f"Output will be written to {out_file_name}")
 
     # Load data into CPU memory since we'll be saving it using CPU libraries
     d = torch.load(args.in_file, map_location=torch.device("cpu"))
@@ -40,20 +73,19 @@ def main():
     # Collect the names of the dicts that can be serialized to JSON; model_state_dict and
     # optimizer_state_dict contain torch.tensors and cannot be JSON serialized
     # In addition, we drop dicts related to training information that is not needed at inference time
-    dicts_to_serialize = ['model_kwargs', 'epoch', 'metadata']
+    dicts_to_serialize = ["model_kwargs", "epoch", "metadata"]
 
-
-    with h5py.File(out_file_name, 'w') as f:
+    with h5py.File(out_file_name, "w") as f:
         # Save small nested dicts as json
-        grp = f.create_group('serialized_dicts')
+        grp = f.create_group("serialized_dicts")
         for k in dicts_to_serialize:
             dict_str = json.dumps(d[k])
             grp.create_dataset(k, data=dict_str)
 
         # Save the OrderedDict containing the model weights
         # The keys are ordered alphanumerically as well.
-        grp_model = f.create_group('model_weights')
-        for k, v in d['model_state_dict'].items():
+        grp_model = f.create_group("model_weights")
+        for k, v in d["model_state_dict"].items():
             if len(v.size()) > 0:
                 grp_model.create_dataset(k, data=v.numpy(), fletcher32=True)
             else:
@@ -66,17 +98,18 @@ def main():
 
         # Metadata for CVMFS LVK distribution
         # This needs to be exactly the same as the "basename" of the hdf5 file
-        f.attrs['CANONICAL_FILE_BASENAME'] = os.path.basename(out_file_name)
+        f.attrs["CANONICAL_FILE_BASENAME"] = os.path.basename(out_file_name)
 
         # Add a few metadata entries as attributes
-        f.attrs['approximant'] = d['metadata']['dataset_settings']['waveform_generator']['approximant']
-        f.attrs['epoch'] = d['epoch']
+        f.attrs["approximant"] = d["metadata"]["dataset_settings"][
+            "waveform_generator"
+        ]["approximant"]
+        f.attrs["epoch"] = d["epoch"]
         # Add the dingo version used for training
-        f.attrs['version'] = str(d.get('version'))
+        f.attrs["version"] = str(d.get("version"))
         # Make it clear to dingo_ls that this is file contains model weights
-        f.attrs['dataset_type'] = 'trained_model' 
+        f.attrs["dataset_type"] = "trained_model"
 
 
 if __name__ == "__main__":
     main()
-
diff --git a/dingo/core/utils/trainutils.py b/dingo/core/utils/trainutils.py
index bc7d12154..a420e8d82 100644
--- a/dingo/core/utils/trainutils.py
+++ b/dingo/core/utils/trainutils.py
@@ -1,10 +1,13 @@
 import time
 import os
+import logging
 import numpy as np
 from os.path import join, isfile
 import csv
 from typing import Literal
 
+log = logging.getLogger(__name__)
+
 
 class AvgTracker:
     def __init__(self):
@@ -85,7 +88,9 @@ def __call__(self, val_loss: float):
         elif score < self.best_score + self.delta:
             self.counter += 1
             if self.verbose:
-                print(f"EarlyStopping counter: {self.counter} out of {self.patience}")
+                log.info(
+                    f"EarlyStopping counter: {self.counter} out of {self.patience}"
+                )
             if self.counter >= self.patience:
                 self.early_stop = True
             return False
@@ -124,23 +129,26 @@ def get_avg(self):
 
     def print_info(self, batch_idx):
         if batch_idx % self.print_freq == 0:
-            print(
-                "{} Epoch: {} [{}/{} ({:.0f}%)]".format(
+            td, td_avg = self.times["Dataloader"].x, self.times["Dataloader"].get_avg()
+            tn, tn_avg = self.times["Network"].x, self.times["Network"].get_avg()
+            log.info(
+                "{} Epoch: {} [{}/{} ({:.0f}%)]\t\t"
+                "Loss: {:.3f} ({:.3f})\t\t"
+                "Time Dataloader: {:.3f} ({:.3f})\t\t"
+                "Time Network: {:.3f} ({:.3f})".format(
                     self.mode,
                     self.epoch,
                     min(batch_idx * self.batch_size, self.len_dataset),
                     self.len_dataset,
                     100.0 * batch_idx * self.batch_size / self.len_dataset,
-                ),
-                end="\t\t",
+                    self.loss,
+                    self.get_avg(),
+                    td,
+                    td_avg,
+                    tn,
+                    tn_avg,
+                )
             )
-            # print loss
-            print(f"Loss: {self.loss:.3f} ({self.get_avg():.3f})", end="\t\t")
-            # print computation times
-            td, td_avg = self.times["Dataloader"].x, self.times["Dataloader"].get_avg()
-            tn, tn_avg = self.times["Network"].x, self.times["Network"].get_avg()
-            print(f"Time Dataloader: {td:.3f} ({td_avg:.3f})", end="\t\t")
-            print(f"Time Network: {tn:.3f} ({tn_avg:.3f})")
 
 
 class RuntimeLimits:
@@ -195,7 +203,7 @@ def limits_exceeded(self, epoch: int = None):
         # check time limit for run
         if self.max_time_per_run is not None:
             if time.time() - self.time_start >= self.max_time_per_run:
-                print(
+                log.info(
                     f"Stop run: Time limit of {self.max_time_per_run} s " f"exceeded."
                 )
                 return True
@@ -204,14 +212,14 @@ def limits_exceeded(self, epoch: int = None):
             if epoch is None:
                 raise ValueError("epoch required")
             if epoch - self.epoch_start >= self.max_epochs_per_run:
-                print(
+                log.info(
                     f"Stop run: Epoch limit of {self.max_epochs_per_run} per run reached."
                 )
                 return True
         # check total epoch limit
         if self.max_epochs_total is not None:
             if epoch >= self.max_epochs_total:
-                print(
+                log.info(
                     f"Stop run: Total epoch limit of {self.max_epochs_total} reached."
                 )
                 return True
@@ -316,13 +324,13 @@ def save_model(pm, log_dir, model_prefix="model", checkpoint_epochs=None):
     """
     # save current model
     model_name = join(log_dir, f"{model_prefix}_latest.pt")
-    print(f"Saving model to {model_name}.", end=" ")
+    log.info(f"Saving model to {model_name}.")
     pm.save_model(model_name, save_training_info=True)
-    print("Done.")
+    log.info("Done.")
 
     # potentially copy model to a checkpoint
     if checkpoint_epochs is not None and pm.epoch % checkpoint_epochs == 0:
         model_name_cp = join(log_dir, f"{model_prefix}_{pm.epoch:03d}.pt")
-        print(f"Copy model to checkpoint {model_name_cp}.", end=" ")
+        log.info(f"Copy model to checkpoint {model_name_cp}.")
         copyfile(model_name, model_name_cp)
-        print("Done.")
+        log.info("Done.")
diff --git a/dingo/gw/SVD.py b/dingo/gw/SVD.py
index e77b7761d..2cad5b40d 100644
--- a/dingo/gw/SVD.py
+++ b/dingo/gw/SVD.py
@@ -1,9 +1,14 @@
+import logging
+
 import numpy as np
 import pandas as pd
 import scipy
 from sklearn.utils.extmath import randomized_svd
 from dingo.core.dataset import DingoDataset
 
+log = logging.getLogger(__name__)
+
+
 class SVDBasis(DingoDataset):
 
     dataset_type = "svd_basis"
@@ -155,15 +160,17 @@ def print_validation_summary(self):
                 if "mismatch" in col:
                     n = int(col.split(sep="=")[-1])
                     mismatches = self.mismatches[col]
-                    print(f"n = {n}")
-                    print("  Mean mismatch = {}".format(np.mean(mismatches)))
-                    print("  Standard deviation = {}".format(np.std(mismatches)))
-                    print("  Max mismatch = {}".format(np.max(mismatches)))
-                    print("  Median mismatch = {}".format(np.median(mismatches)))
-                    print("  Percentiles:")
-                    print("    99    -> {}".format(np.percentile(mismatches, 99)))
-                    print("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-                    print("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
+                    log.info(f"n = {n}")
+                    log.info("  Mean mismatch = {}".format(np.mean(mismatches)))
+                    log.info("  Standard deviation = {}".format(np.std(mismatches)))
+                    log.info("  Max mismatch = {}".format(np.max(mismatches)))
+                    log.info("  Median mismatch = {}".format(np.median(mismatches)))
+                    log.info("  Percentiles:")
+                    log.info("    99    -> {}".format(np.percentile(mismatches, 99)))
+                    log.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
+                    log.info(
+                        "    99.99 -> {}".format(np.percentile(mismatches, 99.99))
+                    )
 
     def decompress(self, coefficients: np.ndarray):
         """
diff --git a/dingo/gw/dataset/evaluate_multibanded_domain.py b/dingo/gw/dataset/evaluate_multibanded_domain.py
index b20a2fb43..07a27d635 100644
--- a/dingo/gw/dataset/evaluate_multibanded_domain.py
+++ b/dingo/gw/dataset/evaluate_multibanded_domain.py
@@ -1,7 +1,9 @@
-import argparse
+import logging
+from typing import Dict
 
+import hydra
 import numpy as np
-import yaml
+from omegaconf import DictConfig, OmegaConf
 from scipy.interpolate import interp1d
 
 from dingo.gw.dataset import generate_parameters_and_polarizations
@@ -14,14 +16,20 @@
     generate_waveforms_parallel,
 )
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
-def _evaluate_multibanding_main(
-    settings_file: str,
+
+def _settings_from_config(cfg: DictConfig) -> Dict:
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    settings.pop("num_samples", None)
+    return settings
+
+
+def evaluate_multibanding(
+    settings: Dict,
     num_samples: int,
 ):
-    with open(settings_file, "r") as f:
-        settings = yaml.safe_load(f)
-
     # Ignore any compression settings
     if "compression" in settings:
         del settings["compression"]
@@ -35,13 +43,13 @@ def _evaluate_multibanding_main(
     settings["intrinsic_prior"]["chirp_mass"] = prior["chirp_mass"].minimum
     # Rebuild prior with updated settings.
     prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    print("Prior")
+    log.info("Prior")
     for k, v in prior.items():
-        print(f"{k}: {v}")
+        log.info(f"{k}: {v}")
 
     domain = build_domain(settings["domain"])
-    print("\nDomain")
-    print(domain.domain_dict)
+    log.info("\nDomain")
+    log.info(domain.domain_dict)
 
     if not isinstance(domain, MultibandedFrequencyDomain):
         raise ValueError("Waveform dataset domain not a MultibandedFrequencyDomain.")
@@ -88,43 +96,34 @@ def _evaluate_multibanding_main(
                 asd_file="aLIGO_ZERO_DET_high_P_asd.txt",
             )
 
-    print("\nMismatches between UFD waveforms and MFD waveforms interpolated to MFD.")
-    print(
+    log.info(
+        "\nMismatches between UFD waveforms and MFD waveforms interpolated to MFD."
+    )
+    log.info(
         "This is a conservative estimate of the MFD performance when training "
         "networks."
     )
     mismatches = np.concatenate([v for v in mismatches.values()])
-    print(f"num_samples = {num_samples}")
-    print("  Mean mismatch = {}".format(np.mean(mismatches)))
-    print("  Standard deviation = {}".format(np.std(mismatches)))
-    print("  Max mismatch = {}".format(np.max(mismatches)))
-    print("  Median mismatch = {}".format(np.median(mismatches)))
-    print("  Percentiles:")
-    print("    99    -> {}".format(np.percentile(mismatches, 99)))
-    print("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-    print("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        description="Evaluate performance of multibanding on waveform dataset.",
-    )
-    parser.add_argument(
-        "--settings-file",
-        type=str,
-        required=True,
-        help="YAML file containing database settings",
-    )
-    parser.add_argument(
-        "--num-samples",
-        type=int,
-        default=5000,
-        help="Number of waveform evaluations for comparison.",
-    )
-    return parser.parse_args()
+    log.info(f"num_samples = {num_samples}")
+    log.info("  Mean mismatch = {}".format(np.mean(mismatches)))
+    log.info("  Standard deviation = {}".format(np.std(mismatches)))
+    log.info("  Max mismatch = {}".format(np.max(mismatches)))
+    log.info("  Median mismatch = {}".format(np.median(mismatches)))
+    log.info("  Percentiles:")
+    log.info("    99    -> {}".format(np.percentile(mismatches, 99)))
+    log.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
+    log.info("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
+
+
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="evaluate_multibanded_domain",
+)
+def main(cfg: DictConfig) -> None:
+    settings = _settings_from_config(cfg)
+    evaluate_multibanding(settings, cfg.num_samples)
 
 
-def main() -> None:
-    args = parse_args()
-    _evaluate_multibanding_main(args.settings_file, args.num_samples)
+if __name__ == "__main__":
+    main()
diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index b35cf49c3..9a5da8c44 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -26,6 +26,7 @@
 from dingo.core.utils.misc import call_func_strict_output_dim
 
 log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
 
 def generate_parameters_and_polarizations(
@@ -262,7 +263,7 @@ def generate_dataset(settings: Dict, num_processes: int) -> WaveformDataset:
 
 @hydra.main(
     version_base="1.3",
-    config_path="../../../configs/examples",
+    config_path="../../../configs",
     config_name="generate_dataset",
 )
 def main(cfg: DictConfig) -> None:
diff --git a/dingo/gw/dataset/utils.py b/dingo/gw/dataset/utils.py
index e8911f87c..4257581a4 100644
--- a/dingo/gw/dataset/utils.py
+++ b/dingo/gw/dataset/utils.py
@@ -1,15 +1,30 @@
+import copy
+import logging
 import argparse
+import sys
 import textwrap
-import copy
+from typing import List
+
 import pandas as pd
 import numpy as np
 import yaml
-from typing import List
 
 from dingo.gw.SVD import SVDBasis
 from dingo.gw.dataset.generate_dataset import train_svd_basis
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
+
+def configure_logging():
+    logging.basicConfig(
+        level=logging.INFO,
+        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
+        stream=sys.stdout,
+        force=True,
+    )
+
 
 def merge_datasets(dataset_list: List[WaveformDataset]) -> WaveformDataset:
     """
@@ -26,7 +41,7 @@ def merge_datasets(dataset_list: List[WaveformDataset]) -> WaveformDataset:
     WaveformDataset containing the merged data.
     """
 
-    print(f"Merging {len(dataset_list)} datasets into one.")
+    log.info(f"Merging {len(dataset_list)} datasets into one.")
 
     # This ensures that all the keys are copied into the new dataset. The "extensive"
     # parts of the dataset (parameters, waveforms) will be overwritten by the combined
@@ -59,14 +74,15 @@ def merge_datasets_cli():
     parallelized waveform generation.
     """
 
+    configure_logging()
     parser = argparse.ArgumentParser(
         formatter_class=argparse.RawDescriptionHelpFormatter,
         description=textwrap.dedent(
             """\
         Combine a collection of datasets into one.
-        
-        Datasets must be in sequentially-labeled HDF5 files with a fixed prefix. 
-        The settings for the new dataset will be based on those of the first file. 
+
+        Datasets must be in sequentially-labeled HDF5 files with a fixed prefix.
+        The settings for the new dataset will be based on those of the first file.
         Optionally, replace the settings with those specified in a YAML file.
         """
         ),
@@ -103,7 +119,7 @@ def merge_datasets_cli():
         merged_dataset.settings = settings
 
     merged_dataset.to_file(args.out_file)
-    print(
+    log.info(
         f"Complete. New dataset consists of {merged_dataset.settings['num_samples']} "
         f"samples."
     )
@@ -114,6 +130,7 @@ def build_svd_cli():
     Command-line function to build an SVD based on an uncompressed dataset file.
     """
 
+    configure_logging()
     parser = argparse.ArgumentParser(
         formatter_class=argparse.RawDescriptionHelpFormatter,
         description=textwrap.dedent(
@@ -137,21 +154,23 @@ def build_svd_cli():
     parser.add_argument(
         "--num_train",
         type=int,
-        help="Number of waveforms to use for training SVD. "
-        "Remainder are used for validation.",
+        help=(
+            "Number of waveforms to use for training SVD. Remainder are used "
+            "for validation."
+        ),
     )
     args = parser.parse_args()
 
     dataset = WaveformDataset(file_name=args.dataset_file)
     if args.num_train is None:
-        n_train = len(WaveformDataset)
+        n_train = len(dataset)
     else:
         n_train = args.num_train
 
     basis, n_train, n_test = train_svd_basis(dataset, args.size, n_train)
     # FIXME: This is not an ideal treatment. We should update the waveform generation
     #  to always provide the requested number of waveforms.
-    print(
+    log.info(
         f"SVD basis trained based on {n_train} waveforms and validated on {n_test} "
         f"waveforms. Note that if this differs from number requested, it will not be "
         f"reflected in the settings file. This is likely due to EOB failure to "
diff --git a/dingo/gw/importance_sampling/diagnostics.py b/dingo/gw/importance_sampling/diagnostics.py
index 39cc95bb5..1733dbcbf 100644
--- a/dingo/gw/importance_sampling/diagnostics.py
+++ b/dingo/gw/importance_sampling/diagnostics.py
@@ -1,4 +1,5 @@
 from os.path import join
+import logging
 import numpy as np
 import math
 import pandas as pd
@@ -6,6 +7,8 @@
 from dingo.core.utils.plotting import plot_corner_multi
 from dingo.gw.result import Result
 
+log = logging.getLogger(__name__)
+
 
 def plot_posterior_slice2d(
     sampler, theta, theta_range, n_grid=100, num_processes=1, outname=None
@@ -219,7 +222,7 @@ def plot_diagnostics(
     inds = np.where(weights > threshold)[0]
     theta_new = theta.loc[inds]
     weights_new = weights[inds]
-    print(
+    log.info(
         f"Generating cornerplot with {len(theta_new)} out of {len(theta)} IS samples."
     )
 
diff --git a/dingo/gw/importance_sampling/importance_weights.py b/dingo/gw/importance_sampling/importance_weights.py
index 7f88fa67e..94e922868 100644
--- a/dingo/gw/importance_sampling/importance_weights.py
+++ b/dingo/gw/importance_sampling/importance_weights.py
@@ -2,53 +2,63 @@
 Step 1: Train unconditional nde
 Step 2: Set up likelihood and prior
 """
+
+import logging
 from pathlib import Path
 
-import yaml
 from os import rename, makedirs
-from os.path import dirname, join, isfile, exists
-import argparse
+from os.path import join, isfile, exists
+
+import hydra
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from dingo.core.posterior_models import NormalizingFlowPosteriorModel
 from dingo.gw.result import Result
 from dingo.gw.inference.gw_samplers import GWSampler
 from dingo.gw.importance_sampling.diagnostics import plot_diagnostics
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
-def parse_args():
-    parser = argparse.ArgumentParser(
-        description="Importance sampling (IS) for dingo models."
-    )
-    parser.add_argument(
-        "--settings",
-        type=str,
-        required=True,
-        help="Path to settings file.",
-    )
-    parser.add_argument(
-        "--outdir",
-        type=str,
-        default=None,
-        help="Output directory for the unconditional nde and IS results.",
-    )
-    args = parser.parse_args()
-
-    if args.outdir is None:
-        args.outdir = dirname(args.settings)
-
-    return args
 
+def _resolve_importance_input_paths(settings: dict) -> None:
+    parameter_samples = settings.get("parameter_samples")
+    if parameter_samples is None:
+        parameter_samples = (
+            settings.get("nde", {}).get("data", {}).get("parameter_samples")
+        )
+    if parameter_samples is None:
+        raise KeyError("Missing required setting parameter_samples.")
 
-def main():
-    # parse args, load settings, load dingo parameter samples
-    args = parse_args()
-    with open(args.settings, "r") as fp:
-        settings = yaml.safe_load(fp)
-    try:
-        result = Result(file_name=settings["parameter_samples"])
-    except KeyError:
-        # except statement for backward compatibility
-        result = Result(file_name=settings["nde"]["data"]["parameter_samples"])
+    parameter_samples = to_absolute_path(parameter_samples)
+    settings["parameter_samples"] = parameter_samples
+    if "nde" in settings:
+        settings["nde"].setdefault("data", {})
+        settings["nde"]["data"]["parameter_samples"] = parameter_samples
+
+    calibration = settings.get("calibration_marginalization")
+    if calibration and "calibration_envelope" in calibration:
+        calibration["calibration_envelope"] = {
+            ifo: to_absolute_path(path)
+            for ifo, path in calibration["calibration_envelope"].items()
+        }
+
+
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="importance_weights",
+)
+def main(cfg: DictConfig):
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    outdir = settings.pop("outdir")
+    if outdir is None:
+        outdir = "."
+    makedirs(outdir, exist_ok=True)
+    _resolve_importance_input_paths(settings)
+
+    result = Result(file_name=settings["parameter_samples"])
     metadata = result.settings
     samples = result.samples
     # for time marginalization, we drop geocent time from the samples
@@ -96,25 +106,23 @@ def main():
     if "log_prob" not in samples.columns:
         # Use GPS time as name for now.
         event_name = str(result.event_metadata["time_event"])
-        nde_name = settings["nde"].get(
-            "path", join(args.outdir, f"nde-{event_name}.pt")
-        )
+        nde_name = settings["nde"].get("path") or join(outdir, f"nde-{event_name}.pt")
         if isfile(nde_name):
-            print(f"Loading nde at {nde_name} for event {event_name}.")
+            log.info(f"Loading nde at {nde_name} for event {event_name}.")
             nde = NormalizingFlowPosteriorModel(
                 model_filename=nde_name,
                 device=settings["nde"]["training"]["device"],
                 load_training_info=False,
             )
         else:
-            print(f"Training new nde for event {event_name}.")
+            log.info(f"Training new nde for event {event_name}.")
             nde = result.train_unconditional_flow(
                 inference_parameters,
                 settings["nde"],
-                train_dir=args.outdir,
+                train_dir=outdir,
             )
-            print(f"Renaming trained nde model to {nde_name}.")
-            rename(join(args.outdir, "model_latest.pt"), nde_name)
+            log.info(f"Renaming trained nde model to {nde_name}.")
+            rename(join(outdir, "model_latest.pt"), nde_name)
 
         # Step 1a: Sample from proposal.
         nde_sampler = GWSampler(model=nde)
@@ -162,10 +170,10 @@ def main():
     # print(np.std(log_evidences) / np.mean(log_evidences_std))
 
     if synthetic_phase:
-        print(f"Sampling synthetic phase.")
+        log.info("Sampling synthetic phase.")
         result.sample_synthetic_phase(synthetic_phase_kwargs)
 
-    print(f"Importance sampling.")
+    log.info("Importance sampling.")
     result.importance_sample(
         num_processes=settings.get("num_processes", 1),
         time_marginalization_kwargs=time_marginalization_kwargs,
@@ -173,13 +181,13 @@ def main():
         calibration_marginalization_kwargs=calibration_marginalization_kwargs,
     )
     result.print_summary()
-    result.to_file(file_name=Path(args.outdir, "dingo_samples_weighted.hdf5"))
+    result.to_file(file_name=Path(outdir, "dingo_samples_weighted.hdf5"))
 
     # Diagnostics
-    diagnostics_dir = join(args.outdir, "IS-diagnostics")
+    diagnostics_dir = join(outdir, "IS-diagnostics")
     if not exists(diagnostics_dir):
         makedirs(diagnostics_dir)
-    print("Plotting diagnostics.")
+    log.info("Plotting diagnostics.")
     plot_diagnostics(
         result,
         diagnostics_dir,
diff --git a/dingo/gw/inference/gw_samplers.py b/dingo/gw/inference/gw_samplers.py
index 2a1b63d89..ecfabf822 100644
--- a/dingo/gw/inference/gw_samplers.py
+++ b/dingo/gw/inference/gw_samplers.py
@@ -1,3 +1,4 @@
+import logging
 from typing import Union, Protocol
 
 import numpy as np
@@ -34,12 +35,13 @@
     MaskDataForFrequencyRangeUpdate,
 )
 
+log = logging.getLogger(__name__)
+
 
 class SamplerProtocol(Protocol):
     base_model_metadata: dict
 
-    def _initialize_transforms(self) -> None:
-        ...
+    def _initialize_transforms(self) -> None: ...
 
 
 class _GWMixinProtocol(SamplerProtocol):
@@ -173,7 +175,6 @@ def _build_domain(self: Sampler):
         if "domain_update" in data_settings:
             self.domain.update(data_settings["domain_update"])
 
-
     def _correct_reference_time(
         self: Sampler, samples: Union[dict, pd.DataFrame], inverse: bool = False
     ):
@@ -240,7 +241,7 @@ def _post_process(self, samples: Union[dict, pd.DataFrame], inverse: bool = Fals
             for k, p in prior.items():
                 if isinstance(p, DeltaFunction) and k not in samples:
                     v = p.peak
-                    print(f"Adding fixed parameter {k} = {v} from prior.")
+                    log.info(f"Adding fixed parameter {k} = {v} from prior.")
                     samples[k] = p.peak * np.ones(num_samples)
         else:
             # Drop non-inference parameters from samples.
@@ -472,8 +473,8 @@ def _initialize_transforms(self):
                 self.gnpe_parameters += transform.input_parameter_names
                 for k, v in transform.kernel.items():
                     self.gnpe_kernel[k] = v
-        print("GNPE parameters: ", self.gnpe_parameters)
-        print("GNPE kernel: ", self.gnpe_kernel)
+        log.info(f"GNPE parameters: {self.gnpe_parameters}")
+        log.info(f"GNPE kernel: {self.gnpe_kernel}")
 
         self.transform_pre = Compose(transform_pre)
 
diff --git a/dingo/gw/ls_cli.py b/dingo/gw/ls_cli.py
index 5c401bee7..6abd3d72e 100644
--- a/dingo/gw/ls_cli.py
+++ b/dingo/gw/ls_cli.py
@@ -1,10 +1,11 @@
 import argparse
 import json
+import logging
+import sys
 from pathlib import Path
-from pprint import pprint
+from pprint import pformat
 
 import h5py
-import torch
 import yaml
 
 from dingo.core.dataset import DingoDataset
@@ -15,19 +16,34 @@
 from dingo.gw.SVD import SVDBasis
 
 
-def ls():
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
+
+def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument("file_name", type=str)
-    args = parser.parse_args()
+    return parser.parse_args()
 
-    path = Path(args.file_name)
+
+def ls():
+    logging.basicConfig(
+        level=logging.INFO,
+        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
+        stream=sys.stdout,
+        force=True,
+    )
+    args = parse_args()
+    file_name = args.file_name
+
+    path = Path(file_name)
     if path.suffix == ".pt":
-        print("Extracting information about torch model.\n")
+        log.info("Extracting information about torch model.\n")
         d, _ = torch_load_with_fallback(path, preferred_map_location="meta")
-        print(f"Version: {d.get('version')}\n")
-        print(f"Model epoch: {d['epoch']}\n")
-        print("Model metadata:")
-        print(
+        log.info(f"Version: {d.get('version')}\n")
+        log.info(f"Model epoch: {d['epoch']}\n")
+        log.info("Model metadata:")
+        log.info(
             yaml.dump(
                 d["metadata"],
                 default_flow_style=False,
@@ -37,14 +53,14 @@ def ls():
 
     elif path.suffix == ".hdf5":
 
-        dataset_type = determine_dataset_type(args.file_name)
+        dataset_type = determine_dataset_type(file_name)
 
         if dataset_type == "gw_result" or dataset_type == "core_result":
-            result = Result(file_name=args.file_name)
-            print(f"Version: {result.version}")
-            print("\nDingo Result\n" + "============\n")
+            result = Result(file_name=file_name)
+            log.info(f"Version: {result.version}")
+            log.info("\nDingo Result\n" + "============\n")
 
-            print(
+            log.info(
                 "Metadata\n"
                 + "--------\n"
                 + yaml.dump(
@@ -54,7 +70,7 @@ def ls():
                 )
             )
             if result.event_metadata:
-                print(
+                log.info(
                     "Event information:\n"
                     + "------------------\n"
                     + yaml.dump(
@@ -64,7 +80,7 @@ def ls():
                     ),
                 )
             if result.importance_sampling_metadata is not None:
-                print(
+                log.info(
                     "Importance sampling:\n"
                     + "--------------------\n"
                     + yaml.dump(
@@ -74,28 +90,28 @@ def ls():
                     ),
                 )
             if result.log_evidence:
-                print("Summary:\n" + "--------")
+                log.info("Summary:\n" + "--------")
                 result.print_summary()
 
         elif dataset_type == "svd_basis":
-            svd = SVDBasis(file_name=args.file_name)
-            print(f"Dingo version: {svd.version}")
-            print("\nSVD Basis\n" + "=========\n")
+            svd = SVDBasis(file_name=file_name)
+            log.info(f"Dingo version: {svd.version}")
+            log.info("\nSVD Basis\n" + "=========\n")
 
-            print(f"Basis size: {svd.n}.")
-            print("\nValidation summary:\n" + "-------------------")
+            log.info(f"Basis size: {svd.n}.")
+            log.info("\nValidation summary:\n" + "-------------------")
             svd.print_validation_summary()
 
         elif dataset_type == "waveform_dataset":
             waveform_dataset = WaveformDataset(
-                file_name=args.file_name, leave_waveforms_on_disk=True
+                file_name=file_name, leave_waveforms_on_disk=True
             )
-            print(f"Dingo version: {waveform_dataset.version}")
-            print("\nWaveform dataset\n" + "================\n")
+            log.info(f"Dingo version: {waveform_dataset.version}")
+            log.info("\nWaveform dataset\n" + "================\n")
 
-            print(f"Dataset size: {len(waveform_dataset)}")
+            log.info(f"Dataset size: {len(waveform_dataset)}")
 
-            print(
+            log.info(
                 "\nSettings\n"
                 + "--------\n"
                 + yaml.dump(
@@ -107,18 +123,18 @@ def ls():
 
             if waveform_dataset.svd:
                 svd = SVDBasis(dictionary=waveform_dataset.svd)
-                print("\nSVD validation summary:\n" + "---------------------------")
+                log.info("\nSVD validation summary:\n" + "---------------------------")
                 svd.print_validation_summary()
 
         elif dataset_type == "asd_dataset":
-            asd_dataset = ASDDataset(file_name=args.file_name)
-            print(f"Dingo version: {asd_dataset.version}")
-            print("\nASD dataset\n" + "================\n")
+            asd_dataset = ASDDataset(file_name=file_name)
+            log.info(f"Dingo version: {asd_dataset.version}")
+            log.info("\nASD dataset\n" + "================\n")
 
-            print(f"Dataset size: {asd_dataset.length_info}\n")
-            print(f"GPS times (min/max): {asd_dataset.gps_info}")
+            log.info(f"Dataset size: {asd_dataset.length_info}\n")
+            log.info(f"GPS times (min/max): {asd_dataset.gps_info}")
 
-            print(
+            log.info(
                 "\nSettings\n"
                 + "--------\n"
                 + yaml.dump(
@@ -129,35 +145,35 @@ def ls():
             )
 
         elif dataset_type == "trained_model":
-            with h5py.File(args.file_name, "r") as f:
-                print("Extracting information about torch model.\n")
-                print(f"Version: {f.attrs['version']}")
-                print(f"Model epoch: {f.attrs['epoch']}")
-                print("Model metadata:")
+            with h5py.File(file_name, "r") as f:
+                log.info("Extracting information about torch model.\n")
+                log.info(f"Version: {f.attrs['version']}")
+                log.info(f"Model epoch: {f.attrs['epoch']}")
+                log.info("Model metadata:")
 
                 for d in ["model_kwargs", "metadata"]:
                     json_data = json.loads(f["serialized_dicts"][d][()])
-                    print(f"\n{d}:\n" + "-" * (len(d) + 1))
-                    pprint(json_data)
+                    log.info(f"\n{d}:\n" + "-" * (len(d) + 1))
+                    log.info(pformat(json_data))
 
         else:
             # Legacy (before dataset_type identifier).
             try:
-                svd = SVDBasis(file_name=args.file_name)
-                print(f"SVD dataset of size n={svd.n}.")
-                print("Validation summary:")
+                svd = SVDBasis(file_name=file_name)
+                log.info(f"SVD dataset of size n={svd.n}.")
+                log.info("Validation summary:")
                 svd.print_validation_summary()
 
             except KeyError:
 
                 dataset = DingoDataset(
-                    file_name=args.file_name,
+                    file_name=file_name,
                     data_keys=[
                         "svd",
                     ],
                 )
                 if dataset.settings is not None:
-                    print(
+                    log.info(
                         yaml.dump(
                             dataset.settings,
                             default_flow_style=False,
@@ -166,13 +182,13 @@ def ls():
                     )
                 if dataset.svd is not None:
                     svd = SVDBasis(dictionary=dataset.svd)
-                    print("SVD validation summary:")
+                    log.info("SVD validation summary:")
                     svd.print_validation_summary()
 
     elif path.suffix == ".yaml":
         with open(path, "r") as f:
             settings = yaml.safe_load(f)
-        print(
+        log.info(
             yaml.dump(
                 settings,
                 default_flow_style=False,
@@ -181,7 +197,7 @@ def ls():
         )
 
     else:
-        print("File type unrecognized.")
+        log.info("File type unrecognized.")
 
 
 def determine_dataset_type(file_name):
diff --git a/dingo/gw/noise/asd_dataset.py b/dingo/gw/noise/asd_dataset.py
index 10e9cf9ef..b4a378f75 100644
--- a/dingo/gw/noise/asd_dataset.py
+++ b/dingo/gw/noise/asd_dataset.py
@@ -1,4 +1,5 @@
 import copy
+import logging
 from pathlib import Path
 from typing import Iterable, Optional
 
@@ -9,6 +10,8 @@
 from dingo.gw.gwutils import *
 from dingo.gw.dataset import DingoDataset
 
+log = logging.getLogger(__name__)
+
 HIGH_ASD_VALUE = 1.0
 
 
@@ -59,7 +62,7 @@ def __init__(
                     self.gps_times.pop(ifo)
 
         if "window_factor" in self.settings["domain_dict"]:
-            print(
+            log.info(
                 "Warning: 'window_factor' is no longer used in ASDDataset. "
                 "Removing from settings."
             )
@@ -135,14 +138,14 @@ def update_domain(self, domain_update):
             self.domain.domain_dict["type"] == "UniformFrequencyDomain"
             and domain_update["type"] == "MultibandedFrequencyDomain"
         ):
-            print("Updating ASD dataset to MultibandedFrequencyDomain.")
+            log.info("Updating ASD dataset to MultibandedFrequencyDomain.")
             asd_dataset_decimated = {}
             mfd = build_domain(domain_update)
             ufd = mfd.base_domain
             if not check_domain_compatibility(self.asds, ufd):
                 # If the ASD length is not compatible with the new base UFD,
                 # first truncate it.
-                print(
+                log.info(
                     f"  Truncating first to new base UniformFrequencyDomain: f_max "
                     f"{self.domain.f_max} Hz -> {ufd.f_max} Hz"
                 )
diff --git a/dingo/gw/noise/asd_estimation.py b/dingo/gw/noise/asd_estimation.py
index 8239b3eab..bc6e9e290 100644
--- a/dingo/gw/noise/asd_estimation.py
+++ b/dingo/gw/noise/asd_estimation.py
@@ -1,10 +1,12 @@
-import argparse
+import logging
 import os
 import pickle
 from os.path import join
 
+import hydra
 import numpy as np
-import yaml
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 from tqdm import tqdm
 
 from dingo.gw.domains import build_domain
@@ -13,6 +15,9 @@
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.utils import psd_data_path
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
 
 def download_and_estimate_psds(
     data_dir: str,
@@ -69,6 +74,11 @@ def download_and_estimate_psds(
     }
     w = get_window(window_kwargs)
     asd_filename_list = {det: [] for det in detectors}
+    total_segments = sum(len(time_segments[det]) for det in detectors)
+    log.info(
+        f"Estimating PSDs for {total_segments} time segments across "
+        f"{len(time_segments)} detector(s)."
+    )
     for det in detectors:
         psd_path = psd_data_path(data_dir, run, det)
         os.makedirs(psd_path, exist_ok=True)
@@ -83,63 +93,54 @@ def download_and_estimate_psds(
         else:
             estimation_kwargs["det"] = det
 
+        log.info(f"Processing {len(time_segments[det])} PSD segment(s) for {det}.")
         for index, (start, end) in enumerate(
             tqdm(time_segments[det], disable=not verbose)
         ):
             filename = join(psd_path, f"asd_{start}.hdf5")
             asd_filename_list[det].append(filename)
-            if not os.path.exists(filename):
-                dataset_dict = {
-                    "settings": {
-                        "dataset_settings": settings["dataset_settings"],
-                        "domain_dict": domain.domain_dict,
-                    }
+            if os.path.exists(filename):
+                log.info(f"ASD file already exists, skipping {filename}.")
+                continue
+
+            log.info(f"Estimating ASD for {det} segment starting at {start}.")
+            dataset_dict = {
+                "settings": {
+                    "dataset_settings": settings["dataset_settings"],
+                    "domain_dict": domain.domain_dict,
                 }
-                psd = estimate_single_psd(time_start=start, **estimation_kwargs)
-                asd = np.sqrt(psd[domain.min_idx : domain.max_idx + 1])
-                gps_time = start
+            }
+            psd = estimate_single_psd(time_start=start, **estimation_kwargs)
+            asd = np.sqrt(psd[domain.min_idx : domain.max_idx + 1])
+            gps_time = start
 
-                dataset_dict["asds"] = {det: np.array([asd])}
-                dataset_dict["gps_times"] = {det: np.array([gps_time])}
+            dataset_dict["asds"] = {det: np.array([asd])}
+            dataset_dict["gps_times"] = {det: np.array([gps_time])}
 
-                dataset = ASDDataset(dictionary=dataset_dict)
-                dataset.to_file(file_name=filename)
+            dataset = ASDDataset(dictionary=dataset_dict)
+            dataset.to_file(file_name=filename)
 
+    log.info("PSD estimation complete.")
     return asd_filename_list
 
 
-def download_and_estimate_cli():
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="estimate_psds",
+)
+def download_and_estimate_cli(cfg: DictConfig):
     """
     Command-line function to download strain data and estimate PSDs based on the data. Used for
     parallelized ASD dataset generation.
     """
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--data_dir",
-        type=str,
-        required=True,
-        help="Path where the PSD data is to be stored.",
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        required=True,
-        help="Path to a settings file containing the settings for the dataset generation",
-    )
-    parser.add_argument(
-        "--time_segments_file",
-        type=str,
-        default=None,
-        help="Path to a file containing the time segments for which PSDs should be estimated",
-    )
-
-    args = parser.parse_args()
 
-    # Load settings
-    with open(args.settings_file, "r") as f:
-        settings = yaml.safe_load(f)
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    data_dir = to_absolute_path(settings.pop("data_dir"))
+    time_segments_file = to_absolute_path(settings.pop("time_segments_file"))
+    verbose = settings.pop("verbose")
 
-    with open(args.time_segments_file, "rb") as f:
+    with open(time_segments_file, "rb") as f:
         time_segments = pickle.load(f)
 
-    download_and_estimate_psds(args.data_dir, settings, time_segments)
+    download_and_estimate_psds(data_dir, settings, time_segments, verbose=verbose)
diff --git a/dingo/gw/noise/generate_dataset.py b/dingo/gw/noise/generate_dataset.py
index e8750e6e2..526a3bd94 100644
--- a/dingo/gw/noise/generate_dataset.py
+++ b/dingo/gw/noise/generate_dataset.py
@@ -1,77 +1,40 @@
-import argparse
-import textwrap
+import logging
 from os.path import join
 
+import hydra
 import os
-import yaml
 import pickle
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from dingo.gw.noise.asd_estimation import (
     download_and_estimate_psds,
 )
 from dingo.gw.noise.asd_dataset import ASDDataset
-from dingo.gw.noise.generate_dataset_dag import create_dag
 from dingo.gw.noise.utils import merge_datasets, get_time_segments
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
-def parse_args():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        description=textwrap.dedent(
-            """\
-        Generate an ASD dataset based on a settings file.
-        """
-        ),
-    )
-    parser.add_argument(
-        "--data_dir",
-        type=str,
-        required=True,
-        help="Path where the PSD data is to be stored. Must contain a 'settings.yaml' file.",
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        default=None,
-        help="Path to a settings file in case two different datasets are generated in the same directory",
-    )
-    parser.add_argument(
-        "--time_segments_file",
-        type=str,
-        default=None,
-        help="Optional file containing a dictionary of a list of time segments that should be used for estimating PSDs."
-        "This has to be a pickle file.",
-    )
-    parser.add_argument(
-        "--out_name",
-        type=str,
-        default=None,
-        help="Path to resulting ASD dataset",
-    )
-    parser.add_argument("--verbose", action="store_true")
 
-    return parser.parse_args()
-
-
-def generate_dataset():
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="generate_asd_dataset",
+)
+def generate_dataset(cfg: DictConfig):
     """
     Creates and saves an ASD dataset
     """
-    args = parse_args()
-
-    # Load settings
-    settings_file = (
-        args.settings_file
-        if args.settings_file is not None
-        else join(args.data_dir, "asd_dataset_settings.yaml")
-    )
-    with open(settings_file, "r") as f:
-        settings = yaml.safe_load(f)
-
-    data_dir = args.data_dir
-
-    if args.time_segments_file:
-        with open(args.time_segments_file, "rb") as f:
+    data_dir = to_absolute_path(cfg.data_dir)
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    time_segments_file = settings.pop("time_segments_file")
+    out_name = settings.pop("out_name")
+    verbose = settings.pop("verbose")
+    settings.pop("data_dir")
+
+    if time_segments_file:
+        with open(to_absolute_path(time_segments_file), "rb") as f:
             time_segments = pickle.load(f)
     else:
         time_segments = get_time_segments(settings["dataset_settings"])
@@ -84,31 +47,37 @@ def generate_dataset():
 
     if "condor" in settings:
 
-        dagman = create_dag(data_dir, settings_file, time_segments, args.out_name)
-
-        try:
-            dagman.visualize(
-                join(data_dir, "tmp", "condor", "ASD_dataset_generation_workflow.png")
-            )
-        except:
-            pass
-
-        dagman.build()
-        print(f"DAG submission file written.")
+        raise NotImplementedError(
+            "Hydra Stage 1 only supports local ASD dataset generation. "
+            "Condor/DAG ASD generation is deferred to Stage 3."
+        )
+        # Legacy Condor/DAG implementation to translate in Stage 3:
+        #
+        # dagman = create_dag(data_dir, settings_file, time_segments, out_name)
+        #
+        # try:
+        #     dagman.visualize(
+        #         join(data_dir, "tmp", "condor", "ASD_dataset_generation_workflow.png")
+        #     )
+        # except:
+        #     pass
+        #
+        # dagman.build()
+        # log.info("DAG submission file written.")
 
     else:
 
-        print("Downloading strain data and estimating PSDs...")
+        log.info("Downloading strain data and estimating PSDs...")
         asd_filename_list = download_and_estimate_psds(
-            args.data_dir, settings, time_segments, verbose=args.verbose
+            data_dir, settings, time_segments, verbose=verbose
         )
         asd_dataset_list = {
             det: [ASDDataset(asd_file) for asd_file in asd_file_list]
             for det, asd_file_list in asd_filename_list.items()
         }
-        print("Merging single dataset files into one...")
+        log.info("Merging single dataset files into one...")
         dataset = merge_datasets(asd_dataset_list)
-        filename = args.out_name
+        filename = out_name
         if filename is None:
             run = settings["dataset_settings"]["observing_run"]
             filename = join(data_dir, f"asds_{run}.hdf5")
diff --git a/dingo/gw/noise/synthetic/asd_sampling.py b/dingo/gw/noise/synthetic/asd_sampling.py
index 707ca586d..1c2f941cb 100644
--- a/dingo/gw/noise/synthetic/asd_sampling.py
+++ b/dingo/gw/noise/synthetic/asd_sampling.py
@@ -1,4 +1,5 @@
 import copy
+import logging
 
 import numpy as np
 from scipy import stats
@@ -8,11 +9,14 @@
     get_index_for_elem,
 )
 
+log = logging.getLogger(__name__)
+
 
 class KDE:
     """
     Kernel Density Estimation (KDE) class for sampling ASDs.
     """
+
     def __init__(self, parameter_dict, sampling_settings):
         """
         Parameters
@@ -53,7 +57,7 @@ def fit(self, weights=None):
                         weights=weights,
                     )
                 except np.linalg.LinAlgError:
-                    print(
+                    log.info(
                         "Warning: Singular Matrix encountered in spectral KDE. Adding small Gaussian noise..."
                     )
                     perturbed_features = spectral_features[:, i, :] + np.random.normal(
@@ -74,7 +78,7 @@ def fit(self, weights=None):
             split_indices = sorted(split_indices)
 
             for i in range(len(split_indices) - 1):
-                vals = y_values[:, split_indices[i]: split_indices[i + 1]].T
+                vals = y_values[:, split_indices[i] : split_indices[i + 1]].T
                 kde_vals = stats.gaussian_kde(
                     vals, bw_method=float(self.settings["bandwidth_spline"])
                 )
@@ -110,9 +114,7 @@ def sample(self, num_samples, rescaling_ys=None):
             # rescale base noise
             if rescaling_ys:
                 y_values_mean = np.mean(y_values, axis=0)
-                y_values = (
-                    y_values - y_values_mean[None, :] + rescaling_ys[det]
-                )
+                y_values = y_values - y_values_mean[None, :] + rescaling_ys[det]
             parameters_dicts[det]["y_values"] = y_values
 
         return parameters_dicts
diff --git a/dingo/gw/noise/synthetic/generate_dataset.py b/dingo/gw/noise/synthetic/generate_dataset.py
index 0f59c17bf..a3896a384 100644
--- a/dingo/gw/noise/synthetic/generate_dataset.py
+++ b/dingo/gw/noise/synthetic/generate_dataset.py
@@ -1,65 +1,24 @@
-import argparse
 import copy
-import textwrap
+import logging
 import numpy as np
 from typing import Dict
 
-import yaml
+import hydra
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.synthetic.asd_parameterization import parameterize_asd_dataset
 from dingo.gw.noise.synthetic.asd_sampling import KDE, get_rescaling_params
 from dingo.gw.noise.synthetic.utils import reconstruct_psds_from_parameters
 
-
-def parse_args():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        description=textwrap.dedent(
-            """\
-        Generate a synthetic noise ASD dataset from an existing dataset of real ASDs. ASDs can be parameterized to generate
-        smooth PSDs, e.g. to obtain a distribution over ASDs similar to BayesWave ASDs, or we can create a dataset over
-        synthetic PSDs to augment the training distribution and enhance robustness. In particular, this allows us to 
-        shift a distribution over ASDs from a source to a target observing run, where insufficient data is available.
-        """
-        ),
-    )
-    parser.add_argument(
-        "--asd_dataset",
-        type=str,
-        required=True,
-        help="Path to existing ASD dataset to be parameterized and re-sampled",
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        required=True,
-        help="YAML file containing database settings",
-    )
-    parser.add_argument(
-        "--num_samples",
-        type=int,
-        default=500,
-        help="Number of samples to draw from the parameterized ASDs",
-    )
-    parser.add_argument(
-        "--num_processes",
-        type=int,
-        default=1,
-        help="Number of processes to use in pool for parallel parameterization",
-    )
-    parser.add_argument(
-        "--out_file",
-        type=str,
-        default="synthetic_asd_dataset.hdf5",
-        help="Name of file for storing dataset.",
-    )
-    parser.add_argument("--verbose", action="store_true")
-
-    return parser.parse_args()
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
 
-def generate_dataset(real_dataset, settings: Dict, num_samples, num_processes: int, verbose: bool):
+def generate_dataset(
+    real_dataset, settings: Dict, num_samples, num_processes: int, verbose: bool
+):
     """
     Generate a synthetic ASD dataset from an existing dataset of real ASDs.
 
@@ -98,7 +57,9 @@ def generate_dataset(real_dataset, settings: Dict, num_samples, num_processes: i
                 settings["parameterization_settings"],
             )
         parameters_dict = kde.sample(num_samples, rescaling_params)
-        synthetic_dataset_dict["settings"]["sampling_settings"] = settings["sampling_settings"]
+        synthetic_dataset_dict["settings"]["sampling_settings"] = settings[
+            "sampling_settings"
+        ]
 
     asds_dict = {}
     for det, params in parameters_dict.items():
@@ -115,24 +76,28 @@ def generate_dataset(real_dataset, settings: Dict, num_samples, num_processes: i
     synthetic_dataset_dict["asd_parameterizations"] = parameters_dict
     synthetic_dataset_dict["asds"] = asds_dict
 
-    return ASDDataset(
-        dictionary=synthetic_dataset_dict
-    )
-
+    return ASDDataset(dictionary=synthetic_dataset_dict)
 
-def main():
-    args = parse_args()
 
-    # Load settings
-    with open(args.settings_file, "r") as f:
-        settings = yaml.safe_load(f)
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../../configs",
+    config_name="generate_synthetic_asd_dataset",
+)
+def main(cfg: DictConfig):
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    asd_dataset = to_absolute_path(settings.pop("asd_dataset"))
+    num_samples = settings.pop("num_samples")
+    num_processes = settings.pop("num_processes")
+    out_file = settings.pop("out_file")
+    verbose = settings.pop("verbose")
 
-    real_dataset = ASDDataset(file_name=args.asd_dataset)
+    real_dataset = ASDDataset(file_name=asd_dataset)
     synthetic_dataset = generate_dataset(
-        real_dataset, settings, args.num_samples, args.num_processes, args.verbose
+        real_dataset, settings, num_samples, num_processes, verbose
     )
 
-    synthetic_dataset.to_file(args.out_file)
+    synthetic_dataset.to_file(out_file)
 
 
 if __name__ == "__main__":
diff --git a/dingo/gw/noise/utils.py b/dingo/gw/noise/utils.py
index 43d1f82ce..dc94d5b44 100644
--- a/dingo/gw/noise/utils.py
+++ b/dingo/gw/noise/utils.py
@@ -1,4 +1,3 @@
-import argparse
 import os.path
 import pickle
 from io import StringIO
@@ -6,14 +5,20 @@
 from os.path import isfile
 import glob
 import copy
+import logging
 
+import hydra
 import numpy as np
 import requests
-import yaml
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from gwpy.table import EventTable
 from dingo.gw.noise.asd_dataset import ASDDataset
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
 """
 Catalogue against which to check that no event is present in the estimated PSDs
 """
@@ -156,7 +161,7 @@ def merge_datasets(asd_dataset_list):
     merged_dict = {"asds": {}, "gps_times": {}}
 
     for det, asd_list in asd_dataset_list.items():
-        print(f"Merging {len(asd_list)} datasets into one for detector {det}.")
+        log.info(f"Merging {len(asd_list)} datasets into one for detector {det}.")
         merged_dict["asds"][det] = np.vstack(
             [asd_dataset.asds[det] for asd_dataset in asd_list]
         )
@@ -171,51 +176,26 @@ def merge_datasets(asd_dataset_list):
     return merged
 
 
-def merge_datasets_cli():
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="merge_asd_datasets",
+)
+def merge_datasets_cli(cfg: DictConfig):
     """
     Command-line function to combine a collection of datasets into one. Used for
     parallelized ASD dataset generation.
     """
 
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--data_dir",
-        type=str,
-        required=True,
-        help="Path where the PSD data is to be stored.",
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        required=True,
-        help="Path to a settings file that contains the settings for the ASD dataset generation",
-    )
-    parser.add_argument(
-        "--time_segments_file",
-        type=str,
-        default=None,
-        help="Path to a file containing the time segments for which PSDs should be estimated",
-    )
-    parser.add_argument(
-        "--num_parts",
-        type=int,
-        default=-1,
-        help="Number of ASD datasets that should be merged",
-    )
-    parser.add_argument(
-        "--out_name",
-        type=str,
-        default=None,
-        help="File name of merged dataset",
-    )
-    args = parser.parse_args()
-
-    with open(args.settings_file, "r") as f:
-        settings = yaml.safe_load(f)
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    data_dir = to_absolute_path(settings.pop("data_dir"))
+    time_segments_file = settings.pop("time_segments_file")
+    num_parts = settings.pop("num_parts")
+    out_name = settings.pop("out_name")
 
     time_segments = None
-    if args.time_segments_file:
-        with open(args.time_segments_file, "rb") as f:
+    if time_segments_file:
+        with open(to_absolute_path(time_segments_file), "rb") as f:
             time_segments = pickle.load(f)
 
     detectors = settings["dataset_settings"]["detectors"]
@@ -223,7 +203,7 @@ def merge_datasets_cli():
     asd_dataset_list = {det: [] for det in detectors}
 
     for det in detectors:
-        file_dir = psd_data_path(args.data_dir, observing_run, det)
+        file_dir = psd_data_path(data_dir, observing_run, det)
         if time_segments:
             filenames = [
                 join(file_dir, f"asd_{seg[0]}.hdf5")
@@ -232,14 +212,16 @@ def merge_datasets_cli():
             ]
         else:  # if no time_segments are specified, use the first 'num_parts' ASD datasets
             filenames = sorted(glob.glob(join(file_dir, f"asd_*.hdf5")))
-            num_parts = min(args.num_parts, len(filenames)) if args.num_parts > 0 else len(filenames)
+            num_parts = (
+                min(num_parts, len(filenames)) if num_parts > 0 else len(filenames)
+            )
             filenames = filenames[:num_parts]
 
         asd_dataset_list[det] = [ASDDataset(filename) for filename in filenames]
 
     merged_dataset = merge_datasets(asd_dataset_list)
-    filename = args.out_name
+    filename = out_name
     if filename is None:
         run = settings["dataset_settings"]["observing_run"]
-        filename = join(args.data_dir, f"asds_{run}.hdf5")
+        filename = join(data_dir, f"asds_{run}.hdf5")
     merged_dataset.to_file(filename)
diff --git a/dingo/gw/result.py b/dingo/gw/result.py
index 51616032d..b5d130ba3 100644
--- a/dingo/gw/result.py
+++ b/dingo/gw/result.py
@@ -1,4 +1,5 @@
 import copy
+import logging
 import time
 from typing import Optional
 
@@ -24,6 +25,7 @@
 
 
 RANDOM_STATE = 150914
+log = logging.getLogger(__name__)
 
 
 class Result(CoreResult):
@@ -199,8 +201,8 @@ def _rebuild_domain(self, verbose=False):
             )
 
             if verbose:
-                print("Rebuilding domain as follows:")
-                print(
+                log.info("Rebuilding domain as follows:")
+                log.info(
                     yaml.dump(
                         domain_dict,
                         default_flow_style=False,
@@ -210,7 +212,7 @@ def _rebuild_domain(self, verbose=False):
             self.domain = build_domain(domain_dict)
         else:
             if verbose:
-                print("No domain updates found; domain not rebuilt.")
+                log.info("No domain updates found; domain not rebuilt.")
 
     def _build_prior(self):
         """Build the prior based on model metadata. Called by __init__()."""
@@ -359,7 +361,7 @@ def _build_likelihood(
         if "updates" in self.importance_sampling_metadata:
             if "T" in self.importance_sampling_metadata["updates"]:
                 delta_f_new = 1 / self.importance_sampling_metadata["updates"]["T"]
-                print(
+                log.info(
                     f'Updating waveform generation delta_f from {wfg_domain_dict["delta_f"]} to {delta_f_new}.'
                 )
                 wfg_domain_dict["delta_f"] = delta_f_new
@@ -415,13 +417,18 @@ def sample_calibration_parameters(self, calibration_sampling_kwargs: dict):
         self.calibration_sampling_kwargs = calibration_sampling_kwargs
 
         # Handle correction_type defaults
-        correction_type = self.calibration_sampling_kwargs.get("correction_type", "data")
+        correction_type = self.calibration_sampling_kwargs.get(
+            "correction_type", "data"
+        )
         if correction_type is None:
             correction_type_dict = {
-                ifo: CALIBRATION_CORRECTION_TYPE_LOOKUP[ifo] for ifo in self.interferometers
+                ifo: CALIBRATION_CORRECTION_TYPE_LOOKUP[ifo]
+                for ifo in self.interferometers
             }
         elif correction_type == "data" or correction_type == "template":
-            correction_type_dict = {ifo: correction_type for ifo in self.interferometers}
+            correction_type_dict = {
+                ifo: correction_type for ifo in self.interferometers
+            }
         elif isinstance(correction_type, dict):
             correction_type_dict = correction_type
         else:
@@ -440,7 +447,7 @@ def sample_calibration_parameters(self, calibration_sampling_kwargs: dict):
             )
 
         # Removing the delta function priors on the frequency nodes, amplitude and phase.
-        # Usually the frequency nodes are set to delta functions, but we also remove the 
+        # Usually the frequency nodes are set to delta functions, but we also remove the
         # the amplitude and phase delta functions if present.
         # This avoids large log probs and log priors, since the density of a delta function
         # at the sampled point is infinite. The delta functions do not affect the sampling,
@@ -452,14 +459,14 @@ def sample_calibration_parameters(self, calibration_sampling_kwargs: dict):
 
         # Sample calibration parameters and calculate log_prob
         num_samples = len(self.samples)
-        print(f"Sampling calibration parameters for {num_samples} samples.")
+        log.info(f"Sampling calibration parameters for {num_samples} samples.")
 
         delta_log_prob = np.zeros(num_samples)
 
         # Here we will sample the calibration parameters from the prior.
-        # We treat the *prior as the proposal* distribution and 
-        # therefore add the log_prob of the sampled calibration parameters 
-        # to the existing log_prob. We also will update the prior 
+        # We treat the *prior as the proposal* distribution and
+        # therefore add the log_prob of the sampled calibration parameters
+        # to the existing log_prob. We also will update the prior
         # to include the calibration priors using the importance_sampling_metadata
         prior_update = self.importance_sampling_metadata.get("prior_update", {})
         for ifo, prior in calibration_priors.items():
@@ -571,7 +578,7 @@ def sample_synthetic_phase(
             self.synthetic_phase_kwargs.get("num_processes", 1), num_valid_samples // 10
         )
 
-        print(f"Estimating synthetic phase for {num_valid_samples} samples.")
+        log.info(f"Estimating synthetic phase for {num_valid_samples} samples.")
         t0 = time.time()
 
         if not inverse:
@@ -660,7 +667,7 @@ def sample_synthetic_phase(
             self.samples["log_prob"] = log_prob_array
             del self.samples["phase"]
 
-        print(f"Done. This took {time.time() - t0:.2f} s.")
+        log.info(f"Done. This took {time.time() - t0:.2f} s.")
 
     def get_samples_bilby_phase(self, num_processes=1):
         """
@@ -690,7 +697,9 @@ def get_samples_bilby_phase(self, num_processes=1):
             num_processes=num_processes,
         )
 
-    def get_pesummary_samples(self, num_processes=1, resampling_method="clip+rejection"):
+    def get_pesummary_samples(
+        self, num_processes=1, resampling_method="clip+rejection"
+    ):
         """Samples in a form suitable for PESummary.
 
         These samples are adjusted to undo certain conventions used internally by
diff --git a/dingo/gw/training/train_builders.py b/dingo/gw/training/train_builders.py
index ca62856bb..f3695015e 100755
--- a/dingo/gw/training/train_builders.py
+++ b/dingo/gw/training/train_builders.py
@@ -1,5 +1,6 @@
 from typing import List, Optional
 import copy
+import logging
 
 import torch.multiprocessing
 import torchvision
@@ -28,6 +29,8 @@
 from dingo.gw.gwutils import *
 from dingo.core.utils import *
 
+log = logging.getLogger(__name__)
+
 
 def build_dataset(
     data_settings: dict,
@@ -82,9 +85,9 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
         List of sub-transforms to omit from the full composition.
     """
 
-    print(f"Setting train transforms.")
+    log.info("Setting train transforms.")
     if omit_transforms is not None:
-        print("Omitting \n\t" + "\n\t".join([t.__name__ for t in omit_transforms]))
+        log.info("Omitting \n\t" + "\n\t".join([t.__name__ for t in omit_transforms]))
 
     # By passing the wfd domain when instantiating the noise dataset, this ensures the
     # domains will match. In particular, it truncates the ASD dataset beyond the new
@@ -142,9 +145,9 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
     # parameters.
     try:
         standardization_dict = data_settings["standardization"]
-        print("Using previously-calculated parameter standardizations.")
+        log.info("Using previously-calculated parameter standardizations.")
     except KeyError:
-        print("Calculating new parameter standardizations.")
+        log.info("Calculating new parameter standardizations.")
         standardization_dict = get_standardization_dict(
             extrinsic_prior_dict,
             wfd,
@@ -258,7 +261,7 @@ def build_svd_for_embedding_network(
         ],
     )
 
-    print("Generating waveforms for embedding network SVD initialization.")
+    log.info("Generating waveforms for embedding network SVD initialization.")
     time_start = time.time()
     ifos = list(wfd[0]["waveform"].keys())
     waveform_len = len(wfd[0]["waveform"][ifos[0]])
@@ -277,7 +280,7 @@ def build_svd_for_embedding_network(
     loader = DataLoader(
         wfd,
         batch_size=batch_size,
-        num_workers= 0,
+        num_workers=0,
         worker_init_fn=fix_random_seeds,
     )
     with threadpool_limits(limits=1, user_api="blas"):
@@ -296,25 +299,25 @@ def build_svd_for_embedding_network(
                 waveforms[ifo][lower : lower + n] = strains[:n]
             if lower + n == num_waveforms:
                 break
-    print(f"...done. This took {time.time() - time_start:.0f} s.")
+    log.info(f"...done. This took {time.time() - time_start:.0f} s.")
 
     # Reset the standard sharing strategy.
     torch.multiprocessing.set_sharing_strategy(old_sharing_strategy)
 
-    print("Generating SVD basis for ifo:")
+    log.info("Generating SVD basis for ifo:")
     time_start = time.time()
     basis_dict = {}
     for ifo in ifos:
         basis = SVDBasis()
         basis.generate_basis(waveforms[ifo][:num_training_samples], size)
         basis_dict[ifo] = basis
-        print(f"...{ifo} done.")
-    print(f"...this took {time.time() - time_start:.0f} s.")
+        log.info(f"...{ifo} done.")
+    log.info(f"...this took {time.time() - time_start:.0f} s.")
 
     if out_dir is not None:
-        print(f"Testing SVD basis matrices.")
+        log.info("Testing SVD basis matrices.")
         for ifo, basis in basis_dict.items():
-            print(f"...{ifo}:")
+            log.info(f"...{ifo}:")
             basis.compute_test_mismatches(
                 waveforms[ifo][num_training_samples:],
                 parameters=parameters.iloc[num_training_samples:].reset_index(
@@ -323,19 +326,19 @@ def build_svd_for_embedding_network(
                 verbose=True,
             )
             basis.to_file(os.path.join(out_dir, f"svd_{ifo}.hdf5"))
-    print("Done")
+    log.info("Done")
 
     # Return V matrices in standard order. Drop the elements below domain.min_idx,
     # since the neural network expects data truncated below these. The dropped elements
     # should be 0.
-    print(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
-    print("...V matrix shapes:")
+    log.info(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
+    log.info("...V matrix shapes:")
     V_rb_list = []
     for ifo in data_settings["detectors"]:
         V = basis_dict[ifo].V
         assert np.allclose(V[: wfd.domain.min_idx], 0)
         V = V[wfd.domain.min_idx :]
-        print("      " + str(V.shape))
+        log.info("      " + str(V.shape))
         V_rb_list.append(V)
-    print("\n")
+    log.info("\n")
     return V_rb_list
diff --git a/dingo/gw/training/train_pipeline.py b/dingo/gw/training/train_pipeline.py
index ac1420b0f..7379e9df5 100644
--- a/dingo/gw/training/train_pipeline.py
+++ b/dingo/gw/training/train_pipeline.py
@@ -1,14 +1,16 @@
 from typing import Optional, Tuple
+import logging
 import os
 
+import hydra
 import numpy as np
 import yaml
-import argparse
 import shutil
-import textwrap
 import time
 from copy import deepcopy
 
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 from threadpoolctl import threadpool_limits
 
 from dingo.core.posterior_models.build_model import (
@@ -30,6 +32,9 @@
 from dingo.gw.dataset import WaveformDataset
 from dingo.core.posterior_models import BasePosteriorModel
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
+
 
 def copy_files_to_local(
     file_path: str,
@@ -61,14 +66,16 @@ def copy_files_to_local(
     if local_dir is not None:
         file_name = file_path.split("/")[-1]
         local_file_path = os.path.join(local_dir, file_name)
-        print(f"Copying file to {local_file_path}")
+        log.info(f"Copying file to {local_file_path}")
         # Copy file
         start_time = time.time()
         shutil.copy(file_path, local_file_path)
         elapsed_time = time.time() - start_time
-        print("Done. This took {:2.0f}:{:2.0f} min.".format(*divmod(elapsed_time, 60)))
+        log.info(
+            "Done. This took {:2.0f}:{:2.0f} min.".format(*divmod(elapsed_time, 60))
+        )
     elif leave_keys_on_disk and is_condor:
-        print(
+        log.info(
             f"Warning: leave_waveforms_on_disk defaults to True, but local_cache_path is not specified. "
             f"This means that the waveforms will be loaded during training from {local_file_path} ."
             f"This can lead to unexpected long times for data loading during training due to network traffic. "
@@ -120,9 +127,10 @@ def prepare_training_new(
     # The embedding network is assumed to have an SVD projection layer. If other types
     # of embedding networks are added in the future, update this code.
 
-    if train_settings["model"].get("embedding_kwargs", None):
+    svd_settings = (train_settings["model"].get("embedding_kwargs") or {}).get("svd")
+    if svd_settings and svd_settings.get("num_training_samples", 0) > 0:
         # First, build the SVD for seeding the embedding network.
-        print("\nBuilding SVD for initialization of embedding network.")
+        log.info("\nBuilding SVD for initialization of embedding network.")
         initial_weights["V_rb_list"] = build_svd_for_embedding_network(
             wfd,
             train_settings["data"],
@@ -153,13 +161,13 @@ def prepare_training_new(
         "train_settings": train_settings,
     }
 
-    print("\nInitializing new posterior model.")
-    print("Complete settings:")
-    print(yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
+    log.info("\nInitializing new posterior model.")
+    log.info("Complete settings:")
+    log.info(yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
 
     pm = build_model_from_kwargs(
         settings=full_settings,
-        initial_weights=initial_weights,
+        initial_weights=initial_weights or None,
         device=local_settings["device"],
     )
 
@@ -173,7 +181,7 @@ def prepare_training_new(
                 **local_settings["wandb"],
             )
         except ImportError:
-            print("WandB is enabled but not installed.")
+            log.info("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -226,7 +234,7 @@ def prepare_training_resume(
                 **local_settings["wandb"],
             )
         except ImportError:
-            print("WandB is enabled but not installed.")
+            log.info("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -278,7 +286,7 @@ def initialize_stage(
     if not resume:
         # New optimizer and scheduler. If we are resuming, these should have been
         # loaded from the checkpoint.
-        print("Initializing new optimizer and scheduler.")
+        log.info("Initializing new optimizer and scheduler.")
         pm.optimizer_kwargs = stage["optimizer"]
         pm.scheduler_kwargs = stage["scheduler"]
         pm.initialize_optimizer_and_scheduler()
@@ -295,7 +303,7 @@ def initialize_stage(
             )
     n_grad = get_number_of_model_parameters(pm.network, (True,))
     n_nograd = get_number_of_model_parameters(pm.network, (False,))
-    print(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}\n")
+    log.info(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}\n")
 
     return train_loader, test_loader
 
@@ -343,14 +351,14 @@ def train_stages(
         stage = stages[n]
 
         if pm.epoch == end_epochs[n] - stage["epochs"]:
-            print(f"\nBeginning training stage {n}. Settings:")
-            print(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            log.info(f"\nBeginning training stage {n}. Settings:")
+            log.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=False
             )
         else:
-            print(f"\nResuming training in stage {n}. Settings:")
-            print(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            log.info(f"\nResuming training in stage {n}. Settings:")
+            log.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=True
             )
@@ -359,7 +367,7 @@ def train_stages(
             try:
                 early_stopping = EarlyStopping(**stage["early_stopping"])
             except Exception:
-                print(
+                log.info(
                     "Early stopping settings invalid. Please pass 'patience', 'delta', 'metric'"
                 )
                 raise
@@ -381,10 +389,10 @@ def train_stages(
 
         if pm.epoch == end_epochs[n]:
             save_file = os.path.join(train_dir, f"model_stage_{n}.pt")
-            print(f"Training stage complete. Saving to {save_file}.")
+            log.info(f"Training stage complete. Saving to {save_file}.")
             pm.save_model(save_file, save_training_info=True)
         if runtime_limits.local_limits_exceeded(pm.epoch):
-            print("Local runtime limits reached. Ending program.")
+            log.info("Local runtime limits reached. Ending program.")
             break
 
     if pm.epoch == end_epochs[-1]:
@@ -393,66 +401,39 @@ def train_stages(
         return False
 
 
-def parse_args():
-    parser = argparse.ArgumentParser(
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        description=textwrap.dedent(
-            """\
-        Train a neural network for gravitational-wave single-event inference.
-        
-        This program can be called in one of two ways:
-            a) with a settings file. This will create a new network based on the 
-            contents of the settings file.
-            b) with a checkpoint file. This will resume training from the checkpoint.
-        """
-        ),
-    )
-    parser.add_argument(
-        "--settings_file",
-        type=str,
-        help="YAML file containing training settings.",
-    )
-    parser.add_argument(
-        "--train_dir", required=True, help="Directory for Dingo training output."
-    )
-    parser.add_argument(
-        "--checkpoint",
-        type=str,
-        help="Checkpoint file from which to resume training.",
+def _resolve_training_input_paths(train_settings: dict) -> None:
+    train_settings["data"]["waveform_dataset_path"] = to_absolute_path(
+        train_settings["data"]["waveform_dataset_path"]
     )
-    parser.add_argument(
-        "--exit_command",
-        type=str,
-        default="",
-        help="Optional command to execute after completion of training.",
-    )
-    args = parser.parse_args()
-
-    # The settings file and checkpoint are mutually exclusive.
-    if args.checkpoint is None and args.settings_file is None:
-        parser.error("Must specify either a checkpoint file or a settings file.")
-    if args.checkpoint is not None and args.settings_file is not None:
-        parser.error("Cannot specify both a checkpoint file and a settings file.")
-
-    return args
+    for stage in train_settings["training"].values():
+        if isinstance(stage, dict) and "asd_dataset_path" in stage:
+            stage["asd_dataset_path"] = to_absolute_path(stage["asd_dataset_path"])
 
 
-def train_local():
-    args = parse_args()
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="train",
+)
+def train_local(cfg: DictConfig):
+    settings = OmegaConf.to_container(cfg, resolve=True)
+    checkpoint = settings.pop("checkpoint")
+    train_dir = settings.pop("train_dir")
+    exit_command = settings.pop("exit_command")
 
-    os.makedirs(args.train_dir, exist_ok=True)
+    os.makedirs(train_dir, exist_ok=True)
 
-    if args.settings_file is not None:
-        print("Beginning new training run.")
-        with open(args.settings_file, "r") as fp:
-            train_settings = yaml.safe_load(fp)
+    if checkpoint is None:
+        log.info("Beginning new training run.")
+        train_settings = settings
+        _resolve_training_input_paths(train_settings)
 
         # Extract the local settings from train settings file, save it separately. This
         # file can later be modified, and the settings take effect immediately upon
         # resuming.
 
         local_settings = train_settings.pop("local")
-        with open(os.path.join(args.train_dir, "local_settings.yaml"), "w") as f:
+        with open(os.path.join(train_dir, "local_settings.yaml"), "w") as f:
             if (
                 local_settings.get("wandb", False)
                 and "id" not in local_settings["wandb"].keys()
@@ -462,31 +443,30 @@ def train_local():
 
                     local_settings["wandb"]["id"] = wandb.util.generate_id()
                 except ImportError:
-                    print("wandb not installed, cannot generate run id.")
+                    log.info("wandb not installed, cannot generate run id.")
             yaml.dump(local_settings, f, default_flow_style=False, sort_keys=False)
 
-        pm, wfd = prepare_training_new(train_settings, args.train_dir, local_settings)
+        pm, wfd = prepare_training_new(train_settings, train_dir, local_settings)
 
     else:
-        if not os.path.isfile(args.checkpoint):
-            raise FileNotFoundError(f"Checkpoint not found: {args.checkpoint}")
-        print("Resuming training run.")
-        with open(os.path.join(args.train_dir, "local_settings.yaml"), "r") as f:
+        checkpoint = to_absolute_path(checkpoint)
+        if not os.path.isfile(checkpoint):
+            raise FileNotFoundError(f"Checkpoint not found: {checkpoint}")
+        log.info("Resuming training run.")
+        with open(os.path.join(train_dir, "local_settings.yaml"), "r") as f:
             local_settings = yaml.safe_load(f)
-        pm, wfd = prepare_training_resume(
-            args.checkpoint, local_settings, args.train_dir
-        )
+        pm, wfd = prepare_training_resume(checkpoint, local_settings, train_dir)
 
     with threadpool_limits(limits=1, user_api="blas"):
-        complete = train_stages(pm, wfd, args.train_dir, local_settings)
+        complete = train_stages(pm, wfd, train_dir, local_settings)
 
     if complete:
-        if args.exit_command:
-            print(
-                f"All training stages complete. Executing exit command: {args.exit_command}."
+        if exit_command:
+            log.info(
+                f"All training stages complete. Executing exit command: {exit_command}."
             )
-            os.system(args.exit_command)
+            os.system(exit_command)
         else:
-            print("All training stages complete.")
+            log.info("All training stages complete.")
     else:
-        print("Program terminated due to runtime limit.")
+        log.info("Program terminated due to runtime limit.")
diff --git a/dingo/gw/training/utils.py b/dingo/gw/training/utils.py
index 4cd0732d8..8e9736280 100644
--- a/dingo/gw/training/utils.py
+++ b/dingo/gw/training/utils.py
@@ -1,23 +1,28 @@
-import argparse
+import logging
 
+import hydra
 import numpy as np
 import torch
 import yaml
+from hydra.utils import to_absolute_path
+from omegaconf import DictConfig, OmegaConf
 
 from dingo.core.utils.backward_compatibility import torch_load_with_fallback
 
+log = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
-def append_stage():
 
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--checkpoint", type=str, required=True)
-    parser.add_argument("--stage_settings_file", type=str, required=True)
-    parser.add_argument("--out_file", type=str, required=True)
-    parser.add_argument("--replace", type=int)
-    args = parser.parse_args()
+@hydra.main(
+    version_base="1.3",
+    config_path="../../../configs",
+    config_name="append_training_stage",
+)
+def append_stage(cfg: DictConfig):
+    cfg = OmegaConf.to_container(cfg, resolve=True)
 
     # trying to load on CUDA, MPS, HIP or CPU
-    d, _ = torch_load_with_fallback(args.checkpoint)
+    d, _ = torch_load_with_fallback(to_absolute_path(cfg["checkpoint"]))
 
     stages = [
         v
@@ -25,34 +30,33 @@ def append_stage():
         if k.startswith("stage_")
     ]
     num_stages = len(stages)
-    print(f"Checkpoint training plan consists of {num_stages} stages.")
+    log.info(f"Checkpoint training plan consists of {num_stages} stages.")
 
-    with open(args.stage_settings_file, "r") as f:
-        new_stage = yaml.safe_load(f)
+    new_stage = cfg["stage"]
 
-    if args.replace is not None:
-        if args.replace < 0 or args.replace >= num_stages:
+    if cfg["replace"] is not None:
+        if cfg["replace"] < 0 or cfg["replace"] >= num_stages:
             raise ValueError(
-                f"Invalid argument replace={args.replace}. Valid values "
+                f"Invalid argument replace={cfg['replace']}. Valid values "
                 f"are {list(range(num_stages))}."
             )
         current_epoch = d["epoch"]
-        stage_epoch = np.sum([s["epochs"] for s in stages[: args.replace]])
+        stage_epoch = np.sum([s["epochs"] for s in stages[: cfg["replace"]]])
         if current_epoch > stage_epoch:
-            print(
+            log.info(
                 f"WARNING: Modification to training plan changes a training stage "
                 f"that has already started. Current model epoch is {current_epoch}. "
                 f"Proceed at your own risk!"
             )
-        print(f"Replacing planned stage {args.replace} with new stage.")
-        new_stage_number = args.replace
+        log.info(f"Replacing planned stage {cfg['replace']} with new stage.")
+        new_stage_number = cfg["replace"]
     else:
-        print(f"Appending new stage to training plan.")
+        log.info("Appending new stage to training plan.")
         new_stage_number = num_stages
 
     d["metadata"]["train_settings"]["training"][f"stage_{new_stage_number}"] = new_stage
-    print("Summary of new training plan:")
-    print(
+    log.info("Summary of new training plan:")
+    log.info(
         yaml.dump(
             d["metadata"]["train_settings"]["training"],
             default_flow_style=False,
@@ -60,4 +64,4 @@ def append_stage():
         )
     )
 
-    torch.save(d, args.out_file)
+    torch.save(d, cfg["out_file"])
diff --git a/dingo/gw/transforms/waveform_transforms.py b/dingo/gw/transforms/waveform_transforms.py
index f6d022435..7bba751ef 100644
--- a/dingo/gw/transforms/waveform_transforms.py
+++ b/dingo/gw/transforms/waveform_transforms.py
@@ -1,8 +1,12 @@
 from typing import Optional
+import logging
+
 import numpy as np
 
 from dingo.gw.domains import MultibandedFrequencyDomain, UniformFrequencyDomain
 
+log = logging.getLogger(__name__)
+
 
 class DecimateAll(object):
     """Transform operator for decimation to multibanded frequency domain."""
@@ -436,14 +440,14 @@ def __init__(
         maximum_frequency: Optional[float | dict[str, float]]
             Update of f_max. If a float, the same value will be used for all detectors.
         print_output: bool
-            Whether to write print statements to the console.
+            Whether to write settings information to the log.
         """
         self.sample_frequencies = domain.sample_frequencies
         self.minimum_frequency = minimum_frequency
         self.maximum_frequency = maximum_frequency
 
         if print_output:
-            print(
+            log.info(
                 f"Transform MaskDataForFrequencyRangeUpdate activated:"
                 f"  Settings: \n"
                 f"    - Minimum_frequency update: {self.minimum_frequency}\n"
diff --git a/dingo/gw/waveform_generator/waveform_generator.py b/dingo/gw/waveform_generator/waveform_generator.py
index f673cf699..078d926a8 100644
--- a/dingo/gw/waveform_generator/waveform_generator.py
+++ b/dingo/gw/waveform_generator/waveform_generator.py
@@ -586,7 +586,7 @@ def generate_FD_waveform(
         # numbers if multibanding is used. If that happens, turn off multibanding to
         # fix this.
         if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-            print(
+            log.info(
                 f"Generation with parameters {parameters_lal} likely numerically "
                 f"unstable due to multibanding, turn off multibanding."
             )
@@ -609,7 +609,7 @@ def generate_FD_waveform(
                 *parameters_lal[lal_dict_idx + 1 :],
             )
             if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-                print(
+                log.info(
                     f"Warning: turning off multibanding for parameters {parameters_lal}"
                     f" likely numerically might not have fixed it, check manually."
                 )
@@ -1624,7 +1624,7 @@ def sum_contributions_m(x_m, phase_shift=0.0):
     pol_m = wfg.generate_hplus_hcross_m(p)
 
     phase_shift = np.random.uniform(high=2 * np.pi)
-    print(f"{phase_shift:.2f}")
+    log.info(f"{phase_shift:.2f}")
     pol = sum_contributions_m(pol_m, phase_shift=phase_shift)
 
     pol_ref = wfg.generate_hplus_hcross({**p, "phase": p["phase"] + phase_shift})
diff --git a/old_examples/__init__.py b/old_examples/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/old_examples/append_training_stage.yaml b/old_examples/append_training_stage.yaml
new file mode 100644
index 000000000..d0e2aa228
--- /dev/null
+++ b/old_examples/append_training_stage.yaml
@@ -0,0 +1,28 @@
+checkpoint: ???
+out_file: updated_model.pt
+replace: null
+
+stage:
+  epochs: 20
+  asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+  freeze_rb_layer: false
+  optimizer:
+    type: adam
+    lr: 1.0e-5
+  scheduler:
+    type: cosine
+    T_max: 20
+  batch_size: 64
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: append_training_stage
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - checkpoint
+          - out_file
+          - stage.asd_dataset_path
diff --git a/old_examples/estimate_psds.yaml b/old_examples/estimate_psds.yaml
new file mode 100644
index 000000000..0a7ed5cbd
--- /dev/null
+++ b/old_examples/estimate_psds.yaml
@@ -0,0 +1,29 @@
+data_dir: asd_dataset
+time_segments_file: ???
+verbose: false
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: estimate_psds
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
diff --git a/old_examples/evaluate_multibanded_domain.yaml b/old_examples/evaluate_multibanded_domain.yaml
new file mode 100644
index 000000000..e4fc40d2d
--- /dev/null
+++ b/old_examples/evaluate_multibanded_domain.yaml
@@ -0,0 +1,45 @@
+# settings for domain of waveforms
+domain:
+  type: MultibandedFrequencyDomain
+  nodes: [20.0, 26.0, 34.0, 46.0, 62.0, 78.0, 1038.0]
+  delta_f_initial: 0.0625
+  base_domain:
+    type: UniformFrequencyDomain
+    f_min: 20.0
+    f_max: 1037.9375
+    delta_f: 0.0625
+
+# settings for waveform generator
+waveform_generator:
+  approximant: IMRPhenomXPHM
+  f_ref: 20.0
+  spin_conversion_phase: 0.0
+
+# Dataset only samples over intrinsic parameters. Extrinsic parameters are chosen at train time.
+intrinsic_prior:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: default
+  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
+  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
+  tilt_1: default
+  tilt_2: default
+  phi_12: default
+  phi_jl: default
+  theta_jn: default
+  # Reference values for fixed (extrinsic) parameters. These are needed to generate a waveform.
+  luminosity_distance: 100.0  # Mpc
+  geocent_time: 0.0  # s
+
+num_samples: 1
+
+compression: null
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: evaluate_multibanded_domain
+    chdir: true
diff --git a/old_examples/generate_asd_dataset.yaml b/old_examples/generate_asd_dataset.yaml
new file mode 100644
index 000000000..6a1bc2327
--- /dev/null
+++ b/old_examples/generate_asd_dataset.yaml
@@ -0,0 +1,31 @@
+data_dir: asd_dataset
+time_segments_file: null
+out_name: null
+verbose: false
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: generate_asd_dataset
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
+          - out_name
diff --git a/configs/examples/generate_dataset.yaml b/old_examples/generate_dataset.yaml
similarity index 100%
rename from configs/examples/generate_dataset.yaml
rename to old_examples/generate_dataset.yaml
diff --git a/old_examples/generate_synthetic_asd_dataset.yaml b/old_examples/generate_synthetic_asd_dataset.yaml
new file mode 100644
index 000000000..be569927d
--- /dev/null
+++ b/old_examples/generate_synthetic_asd_dataset.yaml
@@ -0,0 +1,31 @@
+asd_dataset: ???
+num_samples: 500
+num_processes: 1
+out_file: synthetic_asd_dataset.hdf5
+verbose: false
+
+parameterization_settings:
+  num_spline_positions: 30
+  num_spectral_segments: 400
+  sigma: 0.14
+  delta_f: -1
+  smoothen: true
+
+sampling_settings:
+  bandwidth_spectral: 0.5
+  bandwidth_spline: 0.25
+  split_frequencies:
+    - 30
+    - 100
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: generate_synthetic_asd_dataset
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - asd_dataset
+          - out_file
diff --git a/old_examples/importance_weights.yaml b/old_examples/importance_weights.yaml
new file mode 100644
index 000000000..c1a158d19
--- /dev/null
+++ b/old_examples/importance_weights.yaml
@@ -0,0 +1,57 @@
+outdir: .
+parameter_samples: ???
+num_samples: 1000
+event_dataset: null
+time_marginalization:
+  n_fft: 5
+phase_marginalization: null
+synthetic_phase: null
+calibration_marginalization: null
+num_processes: 1
+
+slice_plots:
+  num_slice_plots: 10
+  params_slice2d:
+    - [phase, geocent_time]
+    - [phase, tilt_1]
+
+nde:
+  path: null
+  data:
+    parameters: null
+  model:
+    posterior_model_type: normalizing_flow
+    posterior_kwargs:
+      num_flow_steps: 10
+      base_transform_kwargs:
+        hidden_dim: 128
+        num_transform_blocks: 2
+        activation: elu
+        dropout_probability: 0.1
+        batch_norm: true
+        num_bins: 8
+        base_transform_type: rq-coupling
+  training:
+    device: cpu
+    num_workers: 0
+    train_fraction: 0.9
+    batch_size: 512
+    epochs: 20
+    optimizer:
+      type: adam
+      lr: 0.003
+    scheduler:
+      type: cosine
+      T_max: 20
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: importance_weights
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - parameter_samples
+          - nde.path
diff --git a/old_examples/merge_asd_datasets.yaml b/old_examples/merge_asd_datasets.yaml
new file mode 100644
index 000000000..289fd535e
--- /dev/null
+++ b/old_examples/merge_asd_datasets.yaml
@@ -0,0 +1,31 @@
+data_dir: asd_dataset
+time_segments_file: null
+num_parts: -1
+out_name: null
+
+dataset_settings:
+  f_s: 4096
+  time_psd: 1024
+  T: 8.0
+  window:
+    roll_off: 0.4
+    type: tukey
+  time_gap: 0
+  num_psds_max: 1
+  detectors:
+    - H1
+    - L1
+  observing_run: O1
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: merge_asd_datasets
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - data_dir
+          - time_segments_file
+          - out_name
diff --git a/old_examples/train.yaml b/old_examples/train.yaml
new file mode 100644
index 000000000..736181fb5
--- /dev/null
+++ b/old_examples/train.yaml
@@ -0,0 +1,84 @@
+checkpoint: null
+train_dir: train
+exit_command: ""
+
+data:
+  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  train_fraction: 0.95
+  detectors:
+    - H1
+    - L1
+  extrinsic_prior:
+    dec: default
+    ra: default
+    geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
+    psi: default
+    luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
+  ref_time: 1126259462.391
+  inference_parameters:
+    - chirp_mass
+    - mass_ratio
+    - chi_1
+    - chi_2
+    - theta_jn
+    - dec
+    - ra
+    - geocent_time
+    - luminosity_distance
+    - psi
+    - phase
+
+model:
+  posterior_model_type: normalizing_flow
+  posterior_kwargs:
+    num_flow_steps: 5
+    base_transform_kwargs:
+      hidden_dim: 64
+      num_transform_blocks: 5
+      activation: elu
+      dropout_probability: 0.0
+      batch_norm: true
+      num_bins: 8
+      base_transform_type: rq-coupling
+  embedding_kwargs:
+    output_dim: 128
+    hidden_dims:
+      - 1024
+      - 512
+      - 256
+      - 128
+    activation: elu
+    dropout: 0.0
+    batch_norm: true
+    svd:
+      num_training_samples: 1000
+      num_validation_samples: 100
+      size: 50
+
+training:
+  stage_0:
+    epochs: 20
+    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    freeze_rb_layer: true
+    optimizer:
+      type: adam
+      lr: 0.0001
+    scheduler:
+      type: cosine
+      T_max: 20
+    batch_size: 64
+
+local:
+  device: cpu
+  num_workers: 6
+  runtime_limits:
+    max_time_per_run: 3600000
+    max_epochs_per_run: 30
+  checkpoint_epochs: 15
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}
+  job:
+    name: train
+    chdir: true
diff --git a/old_examples/unconditional_density_estimation.yaml b/old_examples/unconditional_density_estimation.yaml
new file mode 100644
index 000000000..32f5845d4
--- /dev/null
+++ b/old_examples/unconditional_density_estimation.yaml
@@ -0,0 +1,42 @@
+result_file: ???
+train_dir: nde
+
+data:
+  parameters: null
+
+model:
+  posterior_model_type: normalizing_flow
+  posterior_kwargs:
+    num_flow_steps: 10
+    base_transform_kwargs:
+      hidden_dim: 128
+      num_transform_blocks: 2
+      activation: elu
+      dropout_probability: 0.1
+      batch_norm: true
+      num_bins: 8
+      base_transform_type: rq-coupling
+
+training:
+  device: cpu
+  num_workers: 0
+  train_fraction: 0.9
+  batch_size: 512
+  epochs: 20
+  optimizer:
+    type: adam
+    lr: 0.003
+  scheduler:
+    type: cosine
+    T_max: 20
+
+hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
+  job:
+    name: unconditional_density_estimation
+    chdir: true
+    config:
+      override_dirname:
+        exclude_keys:
+          - result_file
diff --git a/pyproject.toml b/pyproject.toml
index e3341fc16..83bf80f23 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -52,7 +52,7 @@ dependencies = [
 ]
 
 [tool.setuptools.packages.find]
-include = ["dingo*"]
+include = ["dingo*", "configs*"]
 namespaces = false
 
 [tool.setuptools_scm]

From 2c8fecdfd9beaa10d44620d9313249c2682c5b45 Mon Sep 17 00:00:00 2001
From: Heinrich Campe <49278231+hcampe@users.noreply.github.com>
Date: Tue, 7 Jul 2026 16:22:28 +0200
Subject: [PATCH 14/15] started refactoring the code to use hydra and more
 cleanly separate configs and code

---
 HYDRA_TODO.md                                 | 434 ++++++++++++++++--
 configs/append_training_stage.yaml            |   9 +-
 configs/domain/multibanded_frequency.yaml     |   4 +-
 configs/domain/uniform_frequency.yaml         |   2 +-
 configs/experiment/generate_fmpe_dataset.yaml |  24 +-
 configs/experiment/generate_gnpe_dataset.yaml |  24 +-
 configs/experiment/generate_npe_dataset.yaml  |  24 +-
 configs/experiment/train_fmpe.yaml            |  31 +-
 configs/experiment/train_gnpe.yaml            |  87 +++-
 configs/experiment/train_gnpe_init.yaml       |  31 +-
 configs/experiment/train_npe.yaml             |  31 +-
 configs/experiment/train_toy.yaml             |  14 +-
 configs/extrinsic_prior/default.yaml          |  13 +-
 configs/extrinsic_prior/fmpe.yaml             |  13 +-
 configs/hydra/default.yaml                    |   5 +
 configs/intrinsic_prior/default.yaml          |  23 +-
 configs/intrinsic_prior/fmpe.yaml             |  31 +-
 configs/intrinsic_prior/precessing.yaml       |  31 +-
 .../precessing_multibanded_test.yaml          |  31 +-
 configs/model/fmpe.yaml                       |   2 +-
 configs/model/gnpe.yaml                       |   2 +-
 configs/model/gnpe_init.yaml                  |   2 +-
 configs/model/npe.yaml                        |   2 +-
 configs/model/toy_npe.yaml                    |   2 +-
 configs/model/unconditional_npe.yaml          |   2 +-
 configs/optimizer/adam.yaml                   |   3 +-
 configs/scheduler/cosine.yaml                 |   3 +-
 configs/train.yaml                            |  66 ++-
 configs/waveform_generator/phenom_d.yaml      |   1 +
 configs/waveform_generator/phenom_pv2.yaml    |   1 +
 configs/waveform_generator/phenom_xphm.yaml   |   1 +
 dingo/asimov/asimov.py                        |   9 +-
 dingo/core/dataset.py                         |  19 +-
 dingo/core/density/__init__.py                |   1 -
 dingo/core/density/nde_settings.py            |  73 ---
 .../unconditional_density_estimation.py       |  13 +-
 dingo/core/posterior_models/base_model.py     |  54 ++-
 dingo/core/posterior_models/build_model.py    |  65 +--
 dingo/core/posterior_models/cflow_base.py     |   2 +-
 .../core/posterior_models/normalizing_flow.py |   2 +-
 dingo/core/result.py                          |  40 +-
 dingo/core/samplers.py                        |  12 +-
 dingo/core/utils/__init__.py                  |   1 +
 dingo/core/utils/condor_utils.py              |   3 +-
 dingo/core/utils/hydra_utils.py               |  30 ++
 dingo/core/utils/logging_utils.py             |  72 ++-
 dingo/core/utils/pt_to_hdf5.py                |  19 +-
 dingo/core/utils/torchutils.py                |  34 +-
 dingo/core/utils/trainutils.py                |  20 +-
 dingo/gw/SVD.py                               |  20 +-
 dingo/gw/conversion/spin_conversion.py        |   3 +-
 dingo/gw/data/data_download.py                |   5 +-
 dingo/gw/data/data_preparation.py             |   4 +-
 dingo/gw/dataset/compression.py               |  94 ++++
 .../gw/dataset/evaluate_multibanded_domain.py |  76 ++-
 dingo/gw/dataset/generate_dataset.py          | 177 ++-----
 dingo/gw/dataset/generate_dataset_dag.py      |   5 +-
 dingo/gw/dataset/utils.py                     |  30 +-
 dingo/gw/dataset/waveform_dataset.py          |  99 ++--
 dingo/gw/domains/build_domain.py              |  17 +-
 dingo/gw/download_strain_data.py              |   5 +-
 dingo/gw/gwutils.py                           |  18 +-
 dingo/gw/importance_sampling/diagnostics.py   |   4 +-
 .../importance_sampling/importance_weights.py |  14 +-
 dingo/gw/inference/gw_samplers.py             |  40 +-
 dingo/gw/injection.py                         |  19 +-
 dingo/gw/likelihood.py                        |   9 +-
 dingo/gw/ls_cli.py                            |  85 ++--
 dingo/gw/noise/asd_dataset.py                 |   8 +-
 dingo/gw/noise/asd_estimation.py              |  12 +-
 dingo/gw/noise/generate_dataset.py            |   8 +-
 dingo/gw/noise/synthetic/asd_sampling.py      |   4 +-
 dingo/gw/noise/synthetic/generate_dataset.py  |   2 +-
 dingo/gw/noise/utils.py                       |   4 +-
 dingo/gw/prior.py                             |  74 +--
 dingo/gw/result.py                            |  36 +-
 dingo/gw/training/train_builders.py           | 215 +++------
 dingo/gw/training/train_pipeline.py           | 167 +++++--
 dingo/gw/training/train_pipeline_condor.py    |   5 +-
 dingo/gw/training/utils.py                    |  14 +-
 dingo/gw/transforms/noise_transforms.py       |   8 +-
 dingo/gw/transforms/parameter_transforms.py   |   5 +-
 dingo/gw/transforms/waveform_transforms.py    |   6 +-
 .../waveform_generator/waveform_generator.py  |  16 +-
 tests/core/test_posterior_model.py            |  41 +-
 .../test_time_delay_from_geocenter.py         |  10 +-
 tests/gw/test_mfd.py                          |  20 +-
 tests/gw/test_prior_split.py                  |  13 +-
 tests/gw/test_ufd.py                          |  14 +-
 tests/gw/test_waveform_dataset.py             |  97 ++--
 tests/gw/transforms/test_batch_transforms.py  |  14 +-
 .../gw/transforms/test_detector_projection.py |  12 +-
 tests/gw/waveform_generator/test_wfg_m.py     |   4 +-
 tests/gw/waveform_generator/test_wfg_mfd.py   |   4 +-
 tests/test_hydra_config_targets.py            | 260 +++++++++++
 95 files changed, 1995 insertions(+), 1220 deletions(-)
 delete mode 100644 dingo/core/density/nde_settings.py
 create mode 100644 dingo/core/utils/hydra_utils.py
 create mode 100644 dingo/gw/dataset/compression.py
 create mode 100644 tests/test_hydra_config_targets.py

diff --git a/HYDRA_TODO.md b/HYDRA_TODO.md
index f4186cf51..8c4c54fba 100644
--- a/HYDRA_TODO.md
+++ b/HYDRA_TODO.md
@@ -219,8 +219,10 @@ Public console scripts are defined in `pyproject.toml`.
   - Stage 1 has started.
   - Uses Hydra and `instantiate(...)` for some objects already, so it is partly
     beyond a purely superficial Stage 1 pass.
-  - Remaining concern: compression/SVD logic is still procedural and should be
-    revisited in Stage 3.
+  - Stage 3 progress: compression is now represented as an ordered list of
+    Hydra-instantiated transform configs. The train-from-waveforms SVD path is
+    a Hydra target that returns an `SVDBasis`, while the produced SVD arrays are
+    still saved as dataset artifact data.
 
 - [ ] `dingo_generate_dataset_dag`
   - File: `dingo/gw/dataset/generate_dataset_dag.py`
@@ -489,6 +491,142 @@ clear this is a scoped migration rather than an incomplete sweep.
 
 ## Stage 3 Refactor Hotspots
 
+- [ ] Stage 3 implementation philosophy
+  - Prefer moving logic into Hydra configs through defaults, interpolation, and
+    `_target_` entries.
+  - Source-code changes should usually delete old construction/defaulting code
+    or replace it with direct `hydra.utils.instantiate(...)` calls.
+  - Avoid adding new source-code logic. If a new helper seems necessary, discuss
+    it first, except for the already-agreed generic
+    `instantiate_with_runtime_dependencies(...)` helper.
+  - Expect config files to grow and pure Python construction code to shrink.
+  - Start implementation with easier direct-instantiation migrations to
+    establish patterns before touching the transform/metadata-heavy pieces.
+
+- [ ] Define the Stage 3 config-construction contract
+  - Goal: replace old `from_config` / `build_*` / string-dispatch constructors
+    with Hydra-native `_target_` configs and `hydra.utils.instantiate(...)`
+    as the normal construction mechanism.
+  - Target architecture: configs describe the object graph, code defines the
+    objects/functions, and Hydra composes plus instantiates. Avoid config
+    completion helpers, compatibility-shaped builders, and side-channel
+    mutation of settings dictionaries.
+  - Stage 3 implementation rule: only remove old config-construction code or
+    replace it by `_target_` configs plus `hydra.utils.instantiate(...)`.
+    Do not introduce new factories, adapters, compatibility builders, or
+    config-normalization layers without first discussing the design and getting
+    explicit agreement.
+  - Approved helper exception: a small generic
+    `instantiate_with_runtime_dependencies(...)` helper is acceptable for
+    Hydra-instantiating transform lists whose entries sometimes need live
+    runtime objects. This helper should remain generic and should not encode
+    Dingo-specific config completion logic.
+  - Runtime-dependent quantities should be handled by the instantiated objects
+    or by explicit setup methods on those objects, not by patching config
+    dictionaries before construction.
+  - Avoid mixing two construction idioms for the same object family. Once a
+    family is migrated, prefer `_target_` configs over parallel `type` strings
+    plus builder switches.
+  - Difficulty: medium/high because the easy cases are straightforward, but the
+    boundary between declarative construction and normal object lifecycle must
+    be chosen deliberately.
+
+- [ ] Define the saved-metadata contract for Hydra configs
+  - Current datasets, models, and results save settings dictionaries as
+    metadata. Stage 3 must decide what gets saved once configs contain
+    `_target_`, interpolations, defaults composition, and possibly instantiated
+    Python objects.
+  - Candidate policy:
+    - Save the resolved, plain-container snapshot of the composed Hydra config
+      as the authoritative run/model/dataset metadata.
+    - Put as much as possible into the composed config before object
+      construction starts: user choices, defaults, resolved paths, dimensions,
+      standardization settings/statistics, and other values that can be known
+      or computed ahead of time.
+    - All paths saved in metadata should be absolute/resolved paths.
+    - Save the exact config produced by Hydra, rather than translating it into
+      the old `dataset_settings` / `train_settings` metadata shape.
+    - Prefer loading saved config metadata back as an OmegaConf/DictConfig-style
+      object so code can use dot access consistently instead of ordinary nested
+      dictionary indexing.
+    - For results, keep the original resolved model config separate from
+      result-specific metadata. The model config records the proposal/model that
+      produced the samples; event, inference, and importance-sampling metadata
+      record what was done to/with those samples.
+    - Compute parameter standardization before model construction and include
+      the resulting standardization statistics in the resolved config.
+    - Prefer Hydra/OmegaConf composition, interpolation, and resolvers for this
+      ahead-of-time computation over code that mutates config dictionaries
+      during construction.
+    - For simple derived values, use OmegaConf interpolation, e.g. `${...}`, to
+      reference values already present elsewhere in the config.
+    - For heavier ahead-of-time computations, use `_target_` config blocks and
+      `hydra.utils.instantiate(...)` on that config subtree. This is acceptable
+      when the target is the actual computation/object being configured, not a
+      new config-factory layer.
+    - Do not save live instantiated objects; save importable targets,
+      primitive parameters, paths, versions, and artifact provenance instead.
+    - Keep produced arrays/tensors outside the config: model weights,
+      optimizer/scheduler state, SVD basis arrays, result samples, context/event
+      strain arrays, etc. These can remain stored in dedicated artifact/data
+      fields broadly as they are today.
+    - Keep the metadata structure close to the config structure so users can
+      understand saved artifacts by looking at the configs.
+    - Do not design the new metadata around legacy dictionary shapes unless a
+      concrete loader requirement is reintroduced.
+  - If old artifacts need support later, prefer a narrow explicit migration
+    shim at the loading boundary rather than carrying legacy shape through the
+    new config design.
+  - Progress:
+    - HDF5-backed `DingoDataset` settings are now saved as resolved YAML and
+      loaded as OmegaConf/DictConfig metadata, with a fallback for old
+      Python-literal attributes.
+    - Posterior-model metadata is held as DictConfig in memory for dot access,
+      but saved to checkpoints as resolved plain containers so `torch.load`
+      remains compatible with safe loading.
+    - Result-specific metadata remains separate from model metadata as before.
+    - Training computes missing parameter standardizations before the
+      configured training transforms are instantiated, writes the computed
+      statistics back into the training settings, and syncs the resolved
+      transform config that consumes them.
+  - Difficulty: very high. This is a cross-cutting compatibility and
+    reproducibility contract, even if we choose not to preserve old metadata
+    formats.
+
+- [ ] Replace manual config autocompletion/resolution with Hydra-era derived
+      metadata handling
+  - Current hotspot: `dingo/core/posterior_models/build_model.py` contains
+    `autocomplete_model_kwargs(...)`, which mutates model settings in-place
+    based on a runtime data sample.
+  - It currently fills:
+    - `embedding_kwargs.input_dims` from waveform/data shape
+    - `posterior_kwargs.input_dim` from the parameter vector length
+    - `embedding_kwargs.added_context` from whether GNPE proxy context exists
+    - `posterior_kwargs.context_dim` from embedding output dimension plus
+      optional GNPE proxy dimension
+  - This is not plain static config composition: the values depend on the
+    realized training dataset and transform pipeline.
+  - Candidate policy:
+    - Keep user-configurable architecture choices in Hydra configs.
+    - Remove hidden in-place config mutation.
+    - Move any dimensions or settings that can be known before construction into
+      the composed Hydra config, using explicit config values or Hydra
+      resolution mechanisms rather than `autocomplete_model_kwargs(...)`.
+    - Only leave shape inference inside instantiated objects when the value
+      genuinely cannot be known before the object is constructed.
+    - Save the resolved config and any resulting learned/artifact state; do not
+      save a separately patched config unless the object itself owns that state.
+  - Related pipe-only helper: `dingo/pipe/main.py` has
+    `write_complete_config_file(...)`, which writes a completed `.ini` file.
+    Since `.ini` pipe workflows are out of scope for the current Hydra
+    migration, keep this separate for now; if pipe migrates later, Hydra's
+    composed config output should replace most of this behavior.
+  - Difficulty: high. It sits at the intersection of model construction,
+    transform realization, and saved metadata.
+  - Design status: agreed. `autocomplete_model_kwargs(...)` should become
+    superfluous. Ahead-of-time computable dimensions should live in the resolved
+    Hydra config through interpolation or explicit ahead-of-time computation.
+
 - [ ] Move Python-defined defaults into Hydra default configs
   - Goal: make default choices visible, overridable, and composable from the
     config tree rather than hidden in module globals or helper functions.
@@ -503,8 +641,8 @@ clear this is a scoped migration rather than an incomplete sweep.
     - `defaults/importance_sampling`
     - `defaults/nde`
     - `defaults/training`
-  - Keep a deliberate backward-compatibility story for old saved metadata that
-    uses strings such as `default`.
+  - Replace old sentinel values such as `default` with explicit defaults-group
+    choices where possible.
 
 - [ ] Audit implicit constructor/function defaults and expose them in config
   - Example configs currently omit many options because Python functions and
@@ -523,6 +661,8 @@ clear this is a scoped migration rather than an incomplete sweep.
     - transform defaults in the training data pipeline, including
       `zero_noise: false`, optional `random_strain_cropping`, derived
       `context_parameters`, and derived `standardization`
+    - ahead-of-time computed model dimensions, replacing
+      `autocomplete_model_kwargs(...)` where possible
     - pipe/inference defaults currently supplied by parser defaults
     - ASD/noise generation defaults such as frequency range, channels, and
       Condor settings
@@ -532,7 +672,7 @@ clear this is a scoped migration rather than an incomplete sweep.
     - runtime limit defaults (`max_time_per_run`, `max_epochs_per_run`,
       `max_epochs_total`, `epoch_start`)
 
-- [ ] `dingo/core/density/nde_settings.py`
+- [x] `dingo/core/density/nde_settings.py`
   - Move or retire `get_default_nde_settings_3d(...)` and
     `DEFAULT_NDE_SETTINGS_2D`.
   - These currently define additional unconditional NDE architectures:
@@ -540,44 +680,181 @@ clear this is a scoped migration rather than an incomplete sweep.
       Adam learning rate 0.005, cosine `T_max: 10`.
     - 2D GNPE proxy recovery: 5 flow steps, hidden dimension 64, batch size
       4096, 10 epochs, Adam learning rate 0.001, cosine `T_max: 10`.
-  - Decide whether these become separate `model/*` and training config groups,
-    or whether they are superseded by `model/unconditional_npe.yaml`.
+  - Retired in favor of Hydra configs, primarily `model/unconditional_npe.yaml`
+    plus `unconditional_density_estimation.yaml`.
 
 - [ ] `dingo/gw/domains/build_domain.py`
   - Replace `type`-based dispatch with Hydra instantiation.
-  - Keep metadata/backward compatibility in mind.
-
-- [ ] `dingo/gw/prior.py`
+  - New metadata should save the resolved Hydra domain config, not a parallel
+    legacy domain dictionary.
+  - Design status: agreed/easy. Use `_target_` domain configs directly.
+    `build_domain(...)` and `build_domain_from_model_metadata(...)` should
+    become unnecessary once downstream code reads the resolved Hydra config and
+    calls `hydra.utils.instantiate(...)`.
+  - Initial Stage 3 pass: default Hydra domain configs now use `_target_`, and
+    `build_domain(...)` instantiates `_target_` configs while keeping the legacy
+    `type` path for saved metadata and existing callers.
+  - Progress: `dingo_generate_dataset` and
+    `dingo_evaluate_multibanded_domain` now instantiate the Hydra domain config
+    directly instead of calling `build_domain(...)`.
+
+- [x] `dingo/gw/prior.py`
   - Replace `default` string handling and prior builders with explicit config
    targets or reusable config groups.
   - Current code-defined defaults:
     - `default_intrinsic_dict`
     - `default_extrinsic_dict`
     - `default_inference_parameters`
+  - Agreed design:
+    - Remove `default` sentinel strings from the Hydra-era configs.
+    - The intrinsic prior config should target `BBHPriorDict` directly.
+    - The `BBHPriorDict` config should contain explicit entries for the
+      individual parameter priors/constraints.
+    - Prefer making the prior dictionary target explicit over instantiating each
+      Bilby prior object separately, unless a specific parameter requires a
+      different treatment.
+    - Retire `build_prior_with_defaults(...)` once all callers consume the
+      instantiated/configured prior dictionary.
+  - Progress: intrinsic-prior Hydra group configs now target
+    `bilby.gw.prior.BBHPriorDict` directly, with explicit entries replacing
+    `default` sentinels. Extrinsic-prior configs now target
+    `dingo.gw.prior.BBHExtrinsicPriorDict` directly. Dataset generation,
+    multibanded-domain evaluation, training, result, sampler, and injection
+    paths consume these explicit configs. `build_prior_with_defaults(...)`,
+    `get_extrinsic_prior_dict(...)`, and the Python-level default prior /
+    inference-parameter globals were retired. Pipe remains out of scope.
 
 - [ ] Waveform generator construction
   - Replace `new_interface` flags and manual class branching with `_target_`.
+  - Design status: agreed/easy. Use `_target_` waveform-generator configs
+    directly and remove manual branching once callers instantiate from config.
+  - Progress: waveform-generator Hydra group configs now target
+    `dingo.gw.waveform_generator.WaveformGenerator`, and the two dataset
+    entrypoints instantiate them directly with the chosen domain.
 
 - [ ] Compression/SVD workflow
   - Harder than plain instantiation because SVD may be loaded from file or
     trained from generated waveforms.
-  - Possible long-term design: compression is an ordered transform list, with
-    SVD handled by a project-specific factory target.
+  - Possible long-term design: compression is an ordered Hydra-instantiated
+    transform list. An SVD step can be configured as a normal object that knows
+    whether to load an existing basis or train/build one from configured inputs.
+  - Agreed design:
+    - Represent compression as an ordered list of transform configs.
+    - Use different Hydra targets for the different SVD construction paths,
+      e.g. one target that loads an existing `SVDBasis` from file and another
+      target that trains a new `SVDBasis`.
+    - The train-SVD target should return only the `SVDBasis` object needed by
+      `ApplySVD`. Auxiliary information such as actual train/validation counts
+      or mismatch summaries belongs in metadata/provenance, not in the return
+      value used by the transform.
+    - The realized SVD basis should continue to be saved as a produced dataset
+      artifact, embedded in the waveform dataset as today.
+    - The config should save the requested SVD construction method and inputs,
+      not the SVD arrays themselves.
+  - Dataset metadata should save the resolved compression config/request. Any
+    realized SVD arrays remain produced artifact data, saved in the waveform
+    dataset as today.
+  - Progress:
+    - `train_svd_basis(...)` now returns only the `SVDBasis` used by
+      `ApplySVD`.
+    - Dataset-generation compression configs are now ordered lists of
+      Hydra-instantiated transforms.
+    - `train_svd_basis_from_waveforms(...)` is available as a Hydra target for
+      training an `SVDBasis` from generated waveforms and passing it into
+      `ApplySVD`.
+    - The generated SVD arrays remain embedded in the waveform dataset as today.
+    - `WaveformDataset` can reload list-based compression metadata and rebuild
+      decompression transforms.
+    - `load_svd_basis(...)` is available as a Hydra target for loading an
+      existing `SVDBasis` from file and passing it into `ApplySVD`.
+    - Small smoke: `my_runs/compression_svd_smoke` generates and reloads a
+      compressed one-sample dataset with a two-sample SVD-training override.
+  - Remaining work: decide how much SVD training provenance/mismatch
+    information should be saved as metadata.
 
 - [ ] `dingo/gw/training/train_builders.py`
   - Large procedural transform pipeline.
   - Prime candidate for Hydra targets, but many transforms need runtime objects:
     domain, detectors, ASD datasets, standardization, priors, reference time.
-
-- [ ] `dingo/core/posterior_models/build_model.py`
-  - Replace `posterior_model_type` dispatch with `_target_`, while preserving
-    saved-model compatibility.
-
-- [ ] `dingo/core/utils/torchutils.py`
+  - Long-term design option: configs define an ordered transform pipeline with
+    `_target_` entries. Transform objects should receive their dependencies
+    through Hydra-instantiated object graphs or through explicit object setup,
+    not through config mutation.
+  - Agreed runtime-dependency pattern:
+    - Instantiate expensive/shared runtime objects once, e.g. effective domain,
+      waveform dataset, ASD dataset, standardization, and optionally
+      `InterferometerList`.
+    - Package them in a `runtime_dependencies` dictionary.
+    - Instantiate transform configs with a generic helper, conceptually:
+
+      ```python
+      transforms = [
+          hydra.utils.instantiate(t_cfg)(**runtime_dependencies)
+          if t_cfg.get("_partial_", False)
+          else hydra.utils.instantiate(t_cfg)
+          for t_cfg in cfg.transforms
+      ]
+      ```
+
+    - Passing runtime objects this way is not a performance concern because
+      Python passes references, and this happens at transform construction time,
+      not for every sample.
+    - Transforms that consume runtime dependencies should fail early if a
+      required dependency is missing. Avoid silent swallowing of misspelled
+      dependency names.
+    - Do not use fake config interpolations such as `${runtime:asd_dataset}`
+      unless a real resolver is explicitly designed later; Hydra does not
+      provide such a runtime-object resolver by default.
+  - Decide which transform state belongs in config, which belongs inside the
+    instantiated transform/model objects, and which artifact provenance should
+    be saved.
+  - Difficulty: very high. This is probably the hardest non-Condor refactor
+    because the pipeline mixes config, runtime datasets, derived state, and
+    metadata persistence.
+  - Progress: the default training transform order now lives in `configs/train.yaml`
+    as `_target_` entries. `set_train_transforms(...)` instantiates that list via
+    the generic `instantiate_with_runtime_dependencies(...)` helper. Cheap
+    dependencies such as priors, interferometer lists, reference time,
+    parameter dictionaries, standardization, selected output keys, and GNPE
+    transforms are now config-owned and recursively instantiated. The live
+    runtime dependency dictionary is reduced to the waveform domain and ASD
+    dataset. `zero_noise` is handled through the ordinary omit-transform path,
+    and `build_dataset(...)` was retired in favor of instantiating the
+    configured `WaveformDataset`. The standardization-prefix transforms are
+    explicit in config so the old standardization behavior is computed before
+    the final transform list is instantiated.
+
+- [x] `dingo/core/posterior_models/build_model.py`
+  - Replace `posterior_model_type` dispatch with `_target_`.
+  - Remove `autocomplete_model_kwargs(...)` as a config-mutation step. Its
+    responsibilities should either become explicit config fields or normal
+    shape-inference behavior of the instantiated model/embedding objects.
+  - Design status: agreed. Model/checkpoint metadata should save the exact
+    resolved Hydra config plus produced checkpoint state. Old
+    `posterior_model_type` dispatch and config autocompletion should disappear
+    in favor of `_target_` / `instantiate`.
+  - Progress: model configs now carry posterior-model `_target_` entries, the
+    `posterior_model_type` dispatch was removed from the Hydra model path, and
+    `autocomplete_model_kwargs(...)` was retired. Model dimensions are computed
+    before model construction in `prepare_training_new(...)` and saved into the
+    config metadata.
+
+- [x] `dingo/core/utils/torchutils.py`
   - Replace optimizer/scheduler `type` switches with `_target_` configs.
+  - Design status: agreed/easy. Optimizers and schedulers should be direct
+    `_target_` configs, removing the current string-switch helpers.
+  - Initial Stage 3 pass: default optimizer/scheduler configs now use
+    `_target_` partials, and the existing helper functions instantiate those
+    configs while keeping legacy `type` settings for older stage/checkpoint
+    metadata.
+  - Progress: the string-switch fallbacks have been removed from the helper
+    functions, and `BasePosteriorModel.initialize_optimizer_and_scheduler()`
+    now calls `hydra.utils.instantiate(...)` directly.
 
 - [ ] Condor integration as a Hydra workflow/launcher
   - Keep Condor-specific migration out of Stage 1.
+  - Discuss later before implementation. This is intentionally not part of the
+    first Stage 3 implementation pass.
   - Stage 3 goal: compose the job config with Hydra and submit Condor jobs from
     that composed config.
   - Desired user experience: a single Hydra-driven command can configure and
@@ -599,26 +876,113 @@ clear this is a scoped migration rather than an incomplete sweep.
   - These are intentionally deferred while `.ini`/pipe workflows remain outside
     the Stage 1/2 Hydra migration.
 
-- [ ] `dingo/core/density/nde_settings.py`
-  - Move `get_default_nde_settings_3d(...)` and `DEFAULT_NDE_SETTINGS_2D` into
-    Hydra configs.
-  - These can likely become reusable `defaults/nde` entries plus overrides for
-    dimension/device/parameters.
-
 - [ ] `dingo/pipe/parser.py`
   - Biggest migration knot. This is both a CLI parser and a compatibility layer
     with bilby_pipe-style settings.
 
+## Stage 3 Testing Requirements
+
+- [ ] Add extensive tests for Hydra configs and instantiated behavior.
+  - Any option included in the configs should be covered by a test at the
+    appropriate level.
+  - At minimum, every config group/file should have composition tests.
+  - Every `_target_` config should have an instantiation test, unless
+    instantiation is intentionally too expensive; in that case add a cheaper
+    test that validates the config shape and document the skipped expensive
+    behavior.
+  - Behavior-affecting options should have unit or smoke tests that exercise the
+    resulting object behavior, not only config composition.
+  - Important override combinations should be tested, especially for:
+    - domain choices and domain updates
+    - waveform generator variants
+    - prior groups
+    - compression/SVD load-vs-train choices
+    - transform pipeline variants, including GNPE context, zero noise, random
+      cropping, and stage-specific ASD/noise settings
+    - model families and model dimensions
+    - optimizer/scheduler configs
+    - metadata saving/loading with dot-access config restoration
+  - Tests should confirm that saved metadata contains the resolved Hydra config,
+    absolute paths, ahead-of-time computed values such as standardization, and
+    separate artifact/result state.
+  - Progress: `tests/test_hydra_config_targets.py` covers Hydra target
+    composition for domain, waveform generator, priors, models,
+    optimizer/scheduler, SVD load-from-file, standardization-prefix transforms,
+    and an in-memory training-transform smoke that verifies standardization is
+    computed before final transform instantiation.
+
 ## Suggested Order
 
-1. Finish Stage 1 for standalone dataset/noise utilities.
-2. Do Stage 1 for training.
-3. Decide whether `dingo_pipe` should be a compatibility wrapper around Hydra
-   configs or a full Hydra CLI.
-4. Do Stage 2 grouping for dataset generation, ASD generation, and training
-   before touching pipe internals.
-5. Start Stage 3 with the small builder switches:
-   domain, prior, waveform generator, optimizer, scheduler.
-6. Design Condor integration as a Stage 3 Hydra workflow/launcher problem.
-7. Tackle training transforms and model construction.
-8. Leave pipe/parser migration until the rest of the config story is stable.
+1. Start with easy direct `_target_` migrations:
+   domain, waveform generator, optimizer, and scheduler.
+2. Move Python-defined defaults into Hydra config groups and make configs
+   complete. This should be mostly config edits plus deletion of old defaulting
+   code.
+3. Migrate priors: remove `default` sentinels, target `BBHPriorDict` directly,
+   and retire `build_prior_with_defaults(...)`.
+4. Migrate model construction: replace `posterior_model_type` dispatch with
+   `_target_` configs and make `autocomplete_model_kwargs(...)` unnecessary
+   through interpolation / ahead-of-time config computation.
+5. Migrate compression/SVD as an ordered transform list with load-vs-train SVD
+   targets and produced SVD arrays saved as dataset artifacts.
+6. Migrate the training transform pipeline using explicit transform configs and
+   the agreed generic `instantiate_with_runtime_dependencies(...)` helper.
+7. Update metadata saving/loading to save the exact resolved Hydra config,
+   restore dot access on load, keep result metadata separate, and keep produced
+   artifacts in dedicated data/state fields.
+8. Add or expand tests throughout each step; do not leave broad config options
+   untested.
+9. Discuss Condor integration later as a separate Stage 3 workflow/launcher
+   design.
+10. Leave pipe/parser migration until the rest of the config story is stable.
+
+## Stage 3 Risk Notes
+
+These are the areas that need the most care, even if implementation starts with
+the easier direct-instantiation changes.
+
+1. **Saved metadata contract** - very high difficulty.
+   - Cross-cuts datasets, trained models, results, checkpoint resume, and
+     reproducibility.
+   - Needs a deliberate policy for saving resolved Hydra config snapshots that
+     already include all ahead-of-time-computable values, plus artifact
+     provenance and truly produced state without reintroducing parallel legacy
+     metadata dictionaries.
+2. **Training transform pipeline** - very high difficulty.
+   - Many transforms are configured partly by settings and partly by runtime
+     objects: waveform/domain metadata, detectors, ASD datasets, priors,
+     standardization, reference time, and context parameters.
+   - Target design should be a Hydra-instantiated transform/object graph, with
+     runtime setup owned by the objects rather than by config patching.
+3. **Compression/SVD workflow** - high difficulty.
+   - Compression can mean loading an existing basis, training a basis from
+     generated waveforms, applying transforms, and saving enough provenance to
+     reproduce the dataset.
+   - Target design should be a declarative compression-target list of
+     Hydra-instantiated objects, including SVD steps.
+4. **Model construction and checkpoint metadata** - high difficulty.
+   - Replace `posterior_model_type` dispatch with `_target_`.
+   - Model configs also interact strongly with saved metadata.
+5. **Config autocompletion / derived model dimensions** - high difficulty.
+   - `autocomplete_model_kwargs(...)` currently hard-codes model config
+     resolution based on a runtime data sample.
+   - Target design should remove hidden config mutation. Dimensions and related
+     settings should be computed into the resolved Hydra config beforehand
+     whenever possible; object-level inference is only the fallback for values
+     that cannot exist before construction.
+6. **Prior construction and defaults** - medium/high difficulty.
+   - Moving `default` strings to explicit prior config groups is conceptually
+     simple, but Bilby prior objects and constraints make it easy to
+     accidentally change behavior.
+7. **Domain and waveform generator construction** - medium difficulty.
+   - Good candidates for early `_target_` migration once the saved-config
+     policy is clear.
+   - Main risks are waveform-generator interface flags and keeping the config
+     readable.
+8. **Optimizer/scheduler construction** - low/medium difficulty.
+   - Natural `_target_` candidates with relatively small metadata surface.
+   - Good later cleanup once the harder contracts are settled.
+9. **Condor integration** - high difficulty, but separable and deferred.
+   - Important for Stage 3, yet mostly an orchestration/launcher design problem
+     rather than a core config-object construction problem.
+   - Discuss later before implementation.
diff --git a/configs/append_training_stage.yaml b/configs/append_training_stage.yaml
index 88bcd1c58..1dca1ca82 100644
--- a/configs/append_training_stage.yaml
+++ b/configs/append_training_stage.yaml
@@ -12,7 +12,12 @@ stage:
   optimizer: ${optimizer}
   scheduler: ${scheduler}
   epochs: 20
-  asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+  asd_dataset:
+    _target_: dingo.gw.noise.asd_dataset.ASDDataset
+    file_name: training_data/asd_dataset/asds_O1.hdf5
+    ifos: null
+    precision: single
+    domain_update: null
   freeze_rb_layer: false
   batch_size: 64
   early_stopping: null
@@ -25,4 +30,4 @@ hydra:
         exclude_keys:
           - checkpoint
           - out_file
-          - stage.asd_dataset_path
+          - stage.asd_dataset.file_name
diff --git a/configs/domain/multibanded_frequency.yaml b/configs/domain/multibanded_frequency.yaml
index 9b9289d68..2ef881d30 100644
--- a/configs/domain/multibanded_frequency.yaml
+++ b/configs/domain/multibanded_frequency.yaml
@@ -1,8 +1,8 @@
-type: MultibandedFrequencyDomain
+_target_: dingo.gw.domains.MultibandedFrequencyDomain
 nodes: [20.0, 26.0, 34.0, 46.0, 62.0, 78.0, 1038.0]
 delta_f_initial: 0.0625
 base_domain:
-  type: UniformFrequencyDomain
+  _target_: dingo.gw.domains.UniformFrequencyDomain
   f_min: 20.0
   f_max: 1037.9375
   delta_f: 0.0625
diff --git a/configs/domain/uniform_frequency.yaml b/configs/domain/uniform_frequency.yaml
index cb44035cd..3647af448 100644
--- a/configs/domain/uniform_frequency.yaml
+++ b/configs/domain/uniform_frequency.yaml
@@ -1,4 +1,4 @@
-type: UniformFrequencyDomain
+_target_: dingo.gw.domains.UniformFrequencyDomain
 f_min: 20.0
 f_max: 1024.0
 delta_f: 0.125
diff --git a/configs/experiment/generate_fmpe_dataset.yaml b/configs/experiment/generate_fmpe_dataset.yaml
index a62c125fe..308d51fd0 100644
--- a/configs/experiment/generate_fmpe_dataset.yaml
+++ b/configs/experiment/generate_fmpe_dataset.yaml
@@ -8,9 +8,23 @@ defaults:
 
 num_samples: 5000000
 compression:
-  svd:
-    size: 200
-    num_training_samples: 50000
-    num_validation_samples: 10000
-  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+  - _target_: dingo.gw.transforms.WhitenFixedASD
+    _partial_: true
+    _runtime_dependencies_: [domain]
+    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
+    inverse: false
+  - _target_: dingo.gw.SVD.ApplySVD
+    svd_basis:
+      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
+      _partial_: true
+      _runtime_dependencies_:
+        - waveform_generator
+        - prior
+        - num_processes
+        - settings
+        - compression_transforms
+        - compression_settings
+      size: 200
+      num_training_samples: 50000
+      num_validation_samples: 10000
 out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_gnpe_dataset.yaml b/configs/experiment/generate_gnpe_dataset.yaml
index 13cd02db9..4f54a2bb8 100644
--- a/configs/experiment/generate_gnpe_dataset.yaml
+++ b/configs/experiment/generate_gnpe_dataset.yaml
@@ -8,9 +8,23 @@ defaults:
 
 num_samples: 5000000
 compression:
-  svd:
-    size: 200
-    num_training_samples: 50000
-    num_validation_samples: 10000
-  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+  - _target_: dingo.gw.transforms.WhitenFixedASD
+    _partial_: true
+    _runtime_dependencies_: [domain]
+    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
+    inverse: false
+  - _target_: dingo.gw.SVD.ApplySVD
+    svd_basis:
+      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
+      _partial_: true
+      _runtime_dependencies_:
+        - waveform_generator
+        - prior
+        - num_processes
+        - settings
+        - compression_transforms
+        - compression_settings
+      size: 200
+      num_training_samples: 50000
+      num_validation_samples: 10000
 out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_npe_dataset.yaml b/configs/experiment/generate_npe_dataset.yaml
index ebc141c8e..44d952bf8 100644
--- a/configs/experiment/generate_npe_dataset.yaml
+++ b/configs/experiment/generate_npe_dataset.yaml
@@ -8,9 +8,23 @@ defaults:
 
 num_samples: 5000000
 compression:
-  svd:
-    size: 200
-    num_training_samples: 50000
-    num_validation_samples: 10000
-  whitening: aLIGO_ZERO_DET_high_P_asd.txt
+  - _target_: dingo.gw.transforms.WhitenFixedASD
+    _partial_: true
+    _runtime_dependencies_: [domain]
+    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
+    inverse: false
+  - _target_: dingo.gw.SVD.ApplySVD
+    svd_basis:
+      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
+      _partial_: true
+      _runtime_dependencies_:
+        - waveform_generator
+        - prior
+        - num_processes
+        - settings
+        - compression_transforms
+        - compression_settings
+      size: 200
+      num_training_samples: 50000
+      num_validation_samples: 10000
 out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/train_fmpe.yaml b/configs/experiment/train_fmpe.yaml
index 84eaca337..8685b7f99 100644
--- a/configs/experiment/train_fmpe.yaml
+++ b/configs/experiment/train_fmpe.yaml
@@ -6,12 +6,13 @@ defaults:
   - _self_
 
 data:
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    file_name: training_data/waveform_dataset.hdf5
+    domain_update:
+      f_min: 20.0
+      f_max: 512.0
+    svd_size_update: 150
   train_fraction: 0.95
-  domain_update:
-    f_min: 20.0
-    f_max: 512.0
-  svd_size_update: 150
   detectors: [H1, L1]
   ref_time: 1126259462.391
   inference_parameters:
@@ -34,7 +35,12 @@ training:
   evaluate: true
   stage_0:
     epochs: 400
-    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 4096
     early_stopping: null
@@ -44,15 +50,22 @@ training:
       T_max: 400
   stage_1:
     epochs: 100
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 4096
     early_stopping: null
     optimizer:
-      type: adam
+      _target_: torch.optim.Adam
+      _partial_: true
       lr: 1.0e-5
     scheduler:
-      type: cosine
+      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
+      _partial_: true
       T_max: 100
 
 local:
diff --git a/configs/experiment/train_gnpe.yaml b/configs/experiment/train_gnpe.yaml
index d3342ae27..a934225e0 100644
--- a/configs/experiment/train_gnpe.yaml
+++ b/configs/experiment/train_gnpe.yaml
@@ -5,17 +5,29 @@ defaults:
   - _self_
 
 data:
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    file_name: training_data/waveform_dataset.hdf5
+    domain_update:
+      f_min: 20.0
+      f_max: 1024.0
+    svd_size_update: 200
   train_fraction: 0.95
-  domain_update:
-    f_min: 20.0
-    f_max: 1024.0
-  svd_size_update: 200
   detectors: [H1, L1]
   ref_time: 1126259462.391
   gnpe_time_shifts:
+    _target_: dingo.gw.transforms.GNPECoalescenceTimes
+    ifo_list:
+      _target_: bilby.gw.detector.InterferometerList
+      interferometers: ${data.detectors}
     kernel: bilby.core.prior.Uniform(minimum=-0.001, maximum=0.001)
-    exact_equiv: true
+    exact_global_equivariance: true
+    inference: false
+  context_parameters:
+    - L1_time_proxy_relative
+  selected_keys:
+    - inference_parameters
+    - waveform
+    - context_parameters
   inference_parameters:
     - chirp_mass
     - mass_ratio
@@ -31,11 +43,59 @@ data:
     - ra
     - dec
     - psi
+  standardization_transforms:
+    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
+      extrinsic_prior_dict: ${data.extrinsic_prior}
+    - _target_: dingo.gw.transforms.GetDetectorTimes
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+    - ${data.gnpe_time_shifts}
+  transforms:
+    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
+      extrinsic_prior_dict: ${data.extrinsic_prior}
+    - _target_: dingo.gw.transforms.GetDetectorTimes
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+    - ${data.gnpe_time_shifts}
+    - _target_: dingo.gw.transforms.ProjectOntoDetectors
+      _partial_: true
+      _runtime_dependencies_: [domain]
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+    - _target_: dingo.gw.transforms.SampleNoiseASD
+      _partial_: true
+      _runtime_dependencies_: [asd_dataset]
+    - _target_: dingo.gw.transforms.WhitenAndScaleStrain
+      _partial_: true
+      _runtime_dependencies_: [domain]
+    - _target_: dingo.gw.transforms.AddWhiteNoiseComplex
+    - _target_: dingo.gw.transforms.SelectStandardizeRepackageParameters
+      parameters_dict:
+        inference_parameters: ${data.inference_parameters}
+        context_parameters: ${data.context_parameters}
+      standardization_dict: ${data.standardization}
+    - _target_: dingo.gw.transforms.RepackageStrainsAndASDS
+      _partial_: true
+      _runtime_dependencies_: [domain]
+      ifos: ${data.detectors}
+    - _target_: dingo.gw.transforms.UnpackDict
+      selected_keys: ${data.selected_keys}
 
 training:
   stage_0:
     epochs: 300
-    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 4096
     early_stopping: null
@@ -45,15 +105,22 @@ training:
       T_max: 300
   stage_1:
     epochs: 150
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: false
     batch_size: 4096
     early_stopping: null
     optimizer:
-      type: adam
+      _target_: torch.optim.Adam
+      _partial_: true
       lr: 1.0e-5
     scheduler:
-      type: cosine
+      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
+      _partial_: true
       T_max: 150
 
 local:
diff --git a/configs/experiment/train_gnpe_init.yaml b/configs/experiment/train_gnpe_init.yaml
index f47dd2cba..a56de527f 100644
--- a/configs/experiment/train_gnpe_init.yaml
+++ b/configs/experiment/train_gnpe_init.yaml
@@ -5,12 +5,13 @@ defaults:
   - _self_
 
 data:
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    file_name: training_data/waveform_dataset.hdf5
+    domain_update:
+      f_min: 20.0
+      f_max: 1024.0
+    svd_size_update: 200
   train_fraction: 0.95
-  domain_update:
-    f_min: 20.0
-    f_max: 1024.0
-  svd_size_update: 200
   detectors: [H1, L1]
   ref_time: 1126259462.391
   inference_parameters:
@@ -20,7 +21,12 @@ data:
 training:
   stage_0:
     epochs: 150
-    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 4096
     early_stopping: null
@@ -30,15 +36,22 @@ training:
       T_max: 150
   stage_1:
     epochs: 75
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: false
     batch_size: 4096
     early_stopping: null
     optimizer:
-      type: adam
+      _target_: torch.optim.Adam
+      _partial_: true
       lr: 1.0e-5
     scheduler:
-      type: cosine
+      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
+      _partial_: true
       T_max: 75
 
 local:
diff --git a/configs/experiment/train_npe.yaml b/configs/experiment/train_npe.yaml
index 0ecd913ec..f9413befd 100644
--- a/configs/experiment/train_npe.yaml
+++ b/configs/experiment/train_npe.yaml
@@ -5,12 +5,13 @@ defaults:
   - _self_
 
 data:
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    file_name: training_data/waveform_dataset.hdf5
+    domain_update:
+      f_min: 20.0
+      f_max: 1024.0
+    svd_size_update: 200
   train_fraction: 0.95
-  domain_update:
-    f_min: 20.0
-    f_max: 1024.0
-  svd_size_update: 200
   detectors: [H1, L1]
   ref_time: 1126259462.391
   inference_parameters:
@@ -28,7 +29,12 @@ data:
 training:
   stage_0:
     epochs: 300
-    asd_dataset_path: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 4096
     early_stopping: null
@@ -38,15 +44,22 @@ training:
       T_max: 300
   stage_1:
     epochs: 150
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: false
     batch_size: 4096
     early_stopping: null
     optimizer:
-      type: adam
+      _target_: torch.optim.Adam
+      _partial_: true
       lr: 1.0e-5
     scheduler:
-      type: cosine
+      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
+      _partial_: true
       T_max: 150
 
 local:
diff --git a/configs/experiment/train_toy.yaml b/configs/experiment/train_toy.yaml
index 687fad524..a970db3ac 100644
--- a/configs/experiment/train_toy.yaml
+++ b/configs/experiment/train_toy.yaml
@@ -5,10 +5,11 @@ defaults:
   - _self_
 
 data:
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    file_name: training_data/waveform_dataset.hdf5
+    domain_update: null
+    svd_size_update: null
   train_fraction: 0.95
-  domain_update: null
-  svd_size_update: null
   detectors: [H1, L1]
   ref_time: 1126259462.391
   inference_parameters:
@@ -27,7 +28,12 @@ data:
 training:
   stage_0:
     epochs: 20
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 64
     early_stopping: null
diff --git a/configs/extrinsic_prior/default.yaml b/configs/extrinsic_prior/default.yaml
index 702057e6d..25b4a17c7 100644
--- a/configs/extrinsic_prior/default.yaml
+++ b/configs/extrinsic_prior/default.yaml
@@ -1,5 +1,8 @@
-dec: default
-ra: default
-geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
-psi: default
-luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
+_target_: dingo.gw.prior.BBHExtrinsicPriorDict
+_convert_: all
+dictionary:
+  dec: bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')
+  ra: bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")
+  geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
+  psi: bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")
+  luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/extrinsic_prior/fmpe.yaml b/configs/extrinsic_prior/fmpe.yaml
index 4d79f9f7e..ab67e5e28 100644
--- a/configs/extrinsic_prior/fmpe.yaml
+++ b/configs/extrinsic_prior/fmpe.yaml
@@ -1,5 +1,8 @@
-dec: default
-ra: default
-geocent_time: bilby.core.prior.Uniform(minimum=-0.03, maximum=0.03, name='geocent_time')
-psi: default
-luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
+_target_: dingo.gw.prior.BBHExtrinsicPriorDict
+_convert_: all
+dictionary:
+  dec: bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')
+  ra: bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")
+  geocent_time: bilby.core.prior.Uniform(minimum=-0.03, maximum=0.03, name='geocent_time')
+  psi: bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")
+  luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/hydra/default.yaml b/configs/hydra/default.yaml
index 7b9f91d97..f32382a96 100644
--- a/configs/hydra/default.yaml
+++ b/configs/hydra/default.yaml
@@ -4,3 +4,8 @@ hydra:
     dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
   job:
     chdir: true
+  job_logging:
+    loggers:
+      bilby:
+        level: INFO
+        propagate: true
diff --git a/configs/intrinsic_prior/default.yaml b/configs/intrinsic_prior/default.yaml
index 609f3d6e7..c9a697b87 100644
--- a/configs/intrinsic_prior/default.yaml
+++ b/configs/intrinsic_prior/default.yaml
@@ -1,10 +1,13 @@
-mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
-mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
-chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
-mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-phase: default
-chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
-chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
-theta_jn: default
-luminosity_distance: 100.0
-geocent_time: 0.0
+_target_: bilby.gw.prior.BBHPriorDict
+_convert_: all
+dictionary:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
+  chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
+  chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
+  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
+  luminosity_distance: 100.0
+  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/fmpe.yaml b/configs/intrinsic_prior/fmpe.yaml
index 30d74ebb4..41d0f1737 100644
--- a/configs/intrinsic_prior/fmpe.yaml
+++ b/configs/intrinsic_prior/fmpe.yaml
@@ -1,14 +1,17 @@
-mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
-mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
-chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=20.0, maximum=120.0, name='chirp_mass')
-mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-phase: default
-a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
-a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
-tilt_1: default
-tilt_2: default
-phi_12: default
-phi_jl: default
-theta_jn: default
-luminosity_distance: 100.0
-geocent_time: 0.0
+_target_: bilby.gw.prior.BBHPriorDict
+_convert_: all
+dictionary:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=20.0, maximum=120.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
+  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
+  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
+  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
+  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
+  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
+  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
+  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
+  luminosity_distance: 100.0
+  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing.yaml b/configs/intrinsic_prior/precessing.yaml
index 5400d688d..c2648da57 100644
--- a/configs/intrinsic_prior/precessing.yaml
+++ b/configs/intrinsic_prior/precessing.yaml
@@ -1,14 +1,17 @@
-mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
-mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
-chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=150.0, name='chirp_mass')
-mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-phase: default
-a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
-a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
-tilt_1: default
-tilt_2: default
-phi_12: default
-phi_jl: default
-theta_jn: default
-luminosity_distance: 100.0
-geocent_time: 0.0
+_target_: bilby.gw.prior.BBHPriorDict
+_convert_: all
+dictionary:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=150.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
+  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
+  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
+  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
+  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
+  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
+  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
+  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
+  luminosity_distance: 100.0
+  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing_multibanded_test.yaml b/configs/intrinsic_prior/precessing_multibanded_test.yaml
index 7c3f469c1..1db3d7c62 100644
--- a/configs/intrinsic_prior/precessing_multibanded_test.yaml
+++ b/configs/intrinsic_prior/precessing_multibanded_test.yaml
@@ -1,14 +1,17 @@
-mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
-mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
-chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
-mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-phase: default
-a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
-a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
-tilt_1: default
-tilt_2: default
-phi_12: default
-phi_jl: default
-theta_jn: default
-luminosity_distance: 100.0
-geocent_time: 0.0
+_target_: bilby.gw.prior.BBHPriorDict
+_convert_: all
+dictionary:
+  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
+  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
+  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
+  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
+  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
+  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
+  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
+  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
+  luminosity_distance: 100.0
+  geocent_time: 0.0
diff --git a/configs/model/fmpe.yaml b/configs/model/fmpe.yaml
index 94241980e..62d50ced8 100644
--- a/configs/model/fmpe.yaml
+++ b/configs/model/fmpe.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: flow_matching
+_target_: dingo.core.posterior_models.FlowMatchingPosteriorModel
 posterior_kwargs:
   activation: gelu
   batch_norm: true
diff --git a/configs/model/gnpe.yaml b/configs/model/gnpe.yaml
index 17bd48972..a4296b293 100644
--- a/configs/model/gnpe.yaml
+++ b/configs/model/gnpe.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: normalizing_flow
+_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
 posterior_kwargs:
   num_flow_steps: 30
   base_transform_kwargs:
diff --git a/configs/model/gnpe_init.yaml b/configs/model/gnpe_init.yaml
index ee39f8ec4..0a53ebfa5 100644
--- a/configs/model/gnpe_init.yaml
+++ b/configs/model/gnpe_init.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: normalizing_flow
+_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
 posterior_kwargs:
   num_flow_steps: 15
   base_transform_kwargs:
diff --git a/configs/model/npe.yaml b/configs/model/npe.yaml
index 17bd48972..a4296b293 100644
--- a/configs/model/npe.yaml
+++ b/configs/model/npe.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: normalizing_flow
+_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
 posterior_kwargs:
   num_flow_steps: 30
   base_transform_kwargs:
diff --git a/configs/model/toy_npe.yaml b/configs/model/toy_npe.yaml
index 295545298..652707e93 100644
--- a/configs/model/toy_npe.yaml
+++ b/configs/model/toy_npe.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: normalizing_flow
+_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
 posterior_kwargs:
   num_flow_steps: 5
   base_transform_kwargs:
diff --git a/configs/model/unconditional_npe.yaml b/configs/model/unconditional_npe.yaml
index 41c500b77..3dc750652 100644
--- a/configs/model/unconditional_npe.yaml
+++ b/configs/model/unconditional_npe.yaml
@@ -1,4 +1,4 @@
-posterior_model_type: normalizing_flow
+_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
 embedding_kwargs: null
 posterior_kwargs:
   num_flow_steps: 10
diff --git a/configs/optimizer/adam.yaml b/configs/optimizer/adam.yaml
index 30ba48b22..6182e57db 100644
--- a/configs/optimizer/adam.yaml
+++ b/configs/optimizer/adam.yaml
@@ -1,2 +1,3 @@
-type: adam
+_target_: torch.optim.Adam
+_partial_: true
 lr: 0.0001
diff --git a/configs/scheduler/cosine.yaml b/configs/scheduler/cosine.yaml
index c45a236f1..5ba12c5a7 100644
--- a/configs/scheduler/cosine.yaml
+++ b/configs/scheduler/cosine.yaml
@@ -1,2 +1,3 @@
-type: cosine
+_target_: torch.optim.lr_scheduler.CosineAnnealingLR
+_partial_: true
 T_max: 20
diff --git a/configs/train.yaml b/configs/train.yaml
index 631369b33..8ef705247 100644
--- a/configs/train.yaml
+++ b/configs/train.yaml
@@ -12,14 +12,19 @@ exit_command: ""
 
 data:
   extrinsic_prior: ${extrinsic_prior}
-  waveform_dataset_path: training_data/waveform_dataset.hdf5
+  waveform_dataset:
+    _target_: dingo.gw.dataset.WaveformDataset
+    file_name: training_data/waveform_dataset.hdf5
+    precision: single
+    domain_update: null
+    svd_size_update: null
   train_fraction: 0.95
-  domain_update: null
-  svd_size_update: null
   detectors:
     - H1
     - L1
   ref_time: 1126259462.391
+  gnpe_time_shifts: null
+  random_strain_cropping: null
   inference_parameters:
     - chirp_mass
     - mass_ratio
@@ -32,6 +37,54 @@ data:
     - luminosity_distance
     - psi
     - phase
+  context_parameters: []
+  selected_keys:
+    - inference_parameters
+    - waveform
+  standardization:
+    mean: ???
+    std: ???
+  standardization_transforms:
+    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
+      extrinsic_prior_dict: ${data.extrinsic_prior}
+    - _target_: dingo.gw.transforms.GetDetectorTimes
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+  transforms:
+    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
+      extrinsic_prior_dict: ${data.extrinsic_prior}
+    - _target_: dingo.gw.transforms.GetDetectorTimes
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+    - _target_: dingo.gw.transforms.ProjectOntoDetectors
+      _partial_: true
+      _runtime_dependencies_: [domain]
+      ifo_list:
+        _target_: bilby.gw.detector.InterferometerList
+        interferometers: ${data.detectors}
+      ref_time: ${data.ref_time}
+    - _target_: dingo.gw.transforms.SampleNoiseASD
+      _partial_: true
+      _runtime_dependencies_: [asd_dataset]
+    - _target_: dingo.gw.transforms.WhitenAndScaleStrain
+      _partial_: true
+      _runtime_dependencies_: [domain]
+    - _target_: dingo.gw.transforms.AddWhiteNoiseComplex
+    - _target_: dingo.gw.transforms.SelectStandardizeRepackageParameters
+      parameters_dict:
+        inference_parameters: ${data.inference_parameters}
+        context_parameters: ${data.context_parameters}
+      standardization_dict: ${data.standardization}
+    - _target_: dingo.gw.transforms.RepackageStrainsAndASDS
+      _partial_: true
+      _runtime_dependencies_: [domain]
+      ifos: ${data.detectors}
+    - _target_: dingo.gw.transforms.UnpackDict
+      selected_keys: ${data.selected_keys}
 
 training:
   evaluate: false
@@ -39,7 +92,12 @@ training:
     optimizer: ${optimizer}
     scheduler: ${scheduler}
     epochs: 20
-    asd_dataset_path: training_data/asd_dataset/asds_O1.hdf5
+    asd_dataset:
+      _target_: dingo.gw.noise.asd_dataset.ASDDataset
+      file_name: training_data/asd_dataset/asds_O1.hdf5
+      ifos: ${data.detectors}
+      precision: single
+      domain_update: null
     freeze_rb_layer: true
     batch_size: 64
     early_stopping: null
diff --git a/configs/waveform_generator/phenom_d.yaml b/configs/waveform_generator/phenom_d.yaml
index 4ec9d64b3..d56a7d5c7 100644
--- a/configs/waveform_generator/phenom_d.yaml
+++ b/configs/waveform_generator/phenom_d.yaml
@@ -1,3 +1,4 @@
+_target_: dingo.gw.waveform_generator.WaveformGenerator
 approximant: IMRPhenomD
 f_ref: 20.0
 f_start: null
diff --git a/configs/waveform_generator/phenom_pv2.yaml b/configs/waveform_generator/phenom_pv2.yaml
index f4390b2fb..cedd187a6 100644
--- a/configs/waveform_generator/phenom_pv2.yaml
+++ b/configs/waveform_generator/phenom_pv2.yaml
@@ -1,3 +1,4 @@
+_target_: dingo.gw.waveform_generator.WaveformGenerator
 approximant: IMRPhenomPv2
 f_ref: 20.0
 f_start: null
diff --git a/configs/waveform_generator/phenom_xphm.yaml b/configs/waveform_generator/phenom_xphm.yaml
index 8c1a14c17..488f2c656 100644
--- a/configs/waveform_generator/phenom_xphm.yaml
+++ b/configs/waveform_generator/phenom_xphm.yaml
@@ -1,3 +1,4 @@
+_target_: dingo.gw.waveform_generator.WaveformGenerator
 approximant: IMRPhenomXPHM
 f_ref: 20.0
 f_start: null
diff --git a/dingo/asimov/asimov.py b/dingo/asimov/asimov.py
index 230a211d1..9c83d7d3c 100644
--- a/dingo/asimov/asimov.py
+++ b/dingo/asimov/asimov.py
@@ -205,9 +205,9 @@ def collect_logs(self):
                 with open(log, "r") as log_f:
                     message = log_f.read()
                     message = message.split("\n")
-                    messages[log.split("/")[-1]] = "\n".join(message[-100:])
+                    messages[logger.split("/")[-1]] = "\n".join(message[-100:])
             except FileNotFoundError:
-                messages[log.split("/")[-1]] = (
+                messages[logger.split("/")[-1]] = (
                     "There was a problem opening this log file."
                 )
         return messages
@@ -253,7 +253,10 @@ def fmin_max_are_compatible(self, prod_meta, net_meta):
 
     @staticmethod
     def _net_max_luminosity_distance(metadata):
-        prior = metadata["train_settings"]["data"]["extrinsic_prior"]["luminosity_distance"]
+        extrinsic_prior = metadata["train_settings"]["data"]["extrinsic_prior"]
+        prior = extrinsic_prior.get("dictionary", extrinsic_prior)[
+            "luminosity_distance"
+        ]
         match = re.findall(r"maximum=[\d]+", prior)
         assert match
         return int(match[0].split("=")[-1])
diff --git a/dingo/core/dataset.py b/dingo/core/dataset.py
index e15a42fac..45b5c1153 100644
--- a/dingo/core/dataset.py
+++ b/dingo/core/dataset.py
@@ -4,10 +4,12 @@
 import h5py
 import numpy as np
 import pandas as pd
+import yaml
+from omegaconf import OmegaConf
 
 from dingo.core.utils.misc import get_version
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 def recursive_hdf5_save(group, d):
@@ -139,7 +141,7 @@ def __init__(
             self.from_dictionary(dictionary)
 
     def to_file(self, file_name: str, mode: str = "w"):
-        log.info(f"Saving dataset to {file_name}")
+        logger.info(f"Saving dataset to {file_name}")
         save_dict = {
             k: v
             for k, v in vars(self).items()
@@ -148,14 +150,17 @@ def to_file(self, file_name: str, mode: str = "w"):
         with h5py.File(file_name, mode) as f:
             recursive_hdf5_save(f, save_dict)
             if self.settings:
-                f.attrs["settings"] = str(self.settings)
+                f.attrs["settings"] = OmegaConf.to_yaml(
+                    OmegaConf.create(self.settings),
+                    resolve=True,
+                )
             if self.dataset_type:
                 f.attrs["dataset_type"] = self.dataset_type
 
     def from_file(self, file_name: str):
-        log.info(f"Loading dataset from {file_name}.")
+        logger.info(f"Loading dataset from {file_name}.")
         if self._leave_on_disk_keys:
-            log.info(f"Omitting data keys {self._leave_on_disk_keys}.")
+            logger.info(f"Omitting data keys {self._leave_on_disk_keys}.")
 
         with h5py.File(file_name, "r") as f:
             loaded_dict = recursive_hdf5_load(
@@ -167,7 +172,9 @@ def from_file(self, file_name: str):
                 assert k in self._data_keys
                 setattr(self, k, v)
             try:
-                self.settings = ast.literal_eval(f.attrs["settings"])
+                self.settings = OmegaConf.create(yaml.safe_load(f.attrs["settings"]))
+            except yaml.YAMLError:
+                self.settings = OmegaConf.create(ast.literal_eval(f.attrs["settings"]))
             except KeyError:
                 self.settings = None  # Is this necessary?
 
diff --git a/dingo/core/density/__init__.py b/dingo/core/density/__init__.py
index e465636a5..aff991ca2 100644
--- a/dingo/core/density/__init__.py
+++ b/dingo/core/density/__init__.py
@@ -8,4 +8,3 @@
     interpolated_sample_and_log_prob_multi,
     interpolated_log_prob_multi,
 )
-from .nde_settings import get_default_nde_settings_3d
diff --git a/dingo/core/density/nde_settings.py b/dingo/core/density/nde_settings.py
deleted file mode 100644
index e63759843..000000000
--- a/dingo/core/density/nde_settings.py
+++ /dev/null
@@ -1,73 +0,0 @@
-"""Default settings for unconditional density estimation"""
-
-
-def get_default_nde_settings_3d(
-    device="cpu",
-    num_workers=0,
-    inference_parameters=None,
-):
-    settings = {
-        "data": {
-            "inference_parameters": inference_parameters,
-        },
-        "model": {
-            "type": "nsf",
-            "num_flow_steps": 5,
-            "base_transform_kwargs": {
-                "hidden_dim": 128,
-                "num_transform_blocks": 2,
-                "activation": "elu",
-                "dropout_probability": 0.1,
-                "batch_norm": True,
-                "num_bins": 8,
-                "base_transform_type": "rq-coupling",
-            },
-            "input_dim": 3,
-            "context_dim": None,
-        },
-        "training": {
-            "device": device,
-            "num_workers": num_workers,
-            "train_fraction": 0.9,
-            "batch_size": 4096,
-            "epochs": 10,
-            "optimizer": {"type": "adam", "lr": 0.005},
-            "scheduler": {"type": "cosine", "T_max": 10},
-        },
-    }
-    return settings
-
-
-DEFAULT_NDE_SETTINGS_2D = {
-    "data": {
-        "inference_parameters": ["GNPE:H1_time_proxy", "GNPE:L1_time_proxy"],
-        "parameter_samples": None,
-    },
-    "model": {
-        "type": "nsf",
-        "num_flow_steps": 5,
-        "base_transform_kwargs": {
-            "hidden_dim": 64,
-            "num_transform_blocks": 2,
-            "activation": "elu",
-            "dropout_probability": 0.1,
-            "batch_norm": "true",
-            "num_bins": 8,
-            "base_transform_type": "rq-coupling",
-        },
-    },
-    # nde training
-    "training": {
-        "device": "cpu",
-        "num_workers": 0,
-        "train_fraction": 0.9,
-        "batch_size": 4096,
-        "epochs": 10,
-        "optimizer": {"type": "adam"},
-        "lr": 0.001,
-        "scheduler": {
-            "type": "cosine",
-            "T_max": 10,
-        },
-    },
-}
diff --git a/dingo/core/density/unconditional_density_estimation.py b/dingo/core/density/unconditional_density_estimation.py
index 48d056cae..33dfbb5f4 100644
--- a/dingo/core/density/unconditional_density_estimation.py
+++ b/dingo/core/density/unconditional_density_estimation.py
@@ -5,14 +5,13 @@
 import hydra
 import torch
 import numpy as np
-from hydra.utils import to_absolute_path
+from hydra.utils import instantiate, to_absolute_path
 from omegaconf import DictConfig, OmegaConf
 
 from dingo.core.utils import build_train_and_test_loaders
 from dingo.core.utils.trainutils import RuntimeLimits
-from dingo.core.posterior_models import NormalizingFlowPosteriorModel
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -79,8 +78,8 @@ def train_unconditional_density_estimator(
     # set up density estimation network
     settings["model"]["posterior_kwargs"]["input_dim"] = num_params
     settings["model"]["posterior_kwargs"]["context_dim"] = None
-    # TODO: Allow for other types of density estimators (e.g., flow matching).
-    model = NormalizingFlowPosteriorModel(
+    model = instantiate(
+        {"_target_": settings["model"]["_target_"]},
         metadata={"train_settings": settings, "base": copy.deepcopy(result.metadata)},
         device=settings["training"]["device"],
     )
@@ -129,9 +128,9 @@ def main(cfg: DictConfig):
     train_dir = settings.pop("train_dir")
     os.makedirs(train_dir, exist_ok=True)
 
-    log.info(f"Loading result from {result_file}.")
+    logger.info(f"Loading result from {result_file}.")
     result = Result(file_name=result_file)
-    log.info(f"Training unconditional density estimator in {train_dir}.")
+    logger.info(f"Training unconditional density estimator in {train_dir}.")
     train_unconditional_density_estimator(result, settings, train_dir)
 
 
diff --git a/dingo/core/posterior_models/base_model.py b/dingo/core/posterior_models/base_model.py
index 9f0e372cc..30a627841 100755
--- a/dingo/core/posterior_models/base_model.py
+++ b/dingo/core/posterior_models/base_model.py
@@ -11,6 +11,8 @@
 
 import torch
 import dingo.core.utils as utils
+from hydra.utils import instantiate
+from omegaconf import OmegaConf
 from torch.utils.data import Dataset
 import time
 import numpy as np
@@ -24,7 +26,13 @@
 
 from dingo.core.utils.trainutils import EarlyStopping
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
+
+
+def _to_plain_container(value):
+    if OmegaConf.is_config(value):
+        return OmegaConf.to_container(value, resolve=True)
+    return value
 
 
 class BasePosteriorModel(ABC):
@@ -68,7 +76,7 @@ def __init__(
         self.scheduler_kwargs = None
         self.initial_weights = initial_weights
 
-        self.metadata = metadata
+        self.metadata = OmegaConf.create(metadata) if metadata is not None else None
         if self.metadata is not None:
             self.model_kwargs = self.metadata["train_settings"]["model"]
             # Expect self.optimizer_settings and self.scheduler_settings to be set
@@ -201,7 +209,7 @@ def network_to_device(self, device):
         #     raise NotImplementedError('This needs testing!')
         #     # dim = 0 [512, ...] -> [256, ...], [256, ...] on 2 GPUs
         #     self.network = torch.nn.DataParallel(self.network)
-        log.info(f"Putting posterior model to device {self.device}.")
+        logger.info(f"Putting posterior model to device {self.device}.")
         self.network.to(self.device)
 
     def initialize_optimizer_and_scheduler(self):
@@ -210,12 +218,18 @@ def initialize_optimizer_and_scheduler(self):
         and self.scheduler_kwargs, respectively.
         """
         if self.optimizer_kwargs is not None:
-            self.optimizer = utils.get_optimizer_from_kwargs(
-                self.network.parameters(), **self.optimizer_kwargs
+            optimizer = instantiate(self.optimizer_kwargs)
+            self.optimizer = (
+                optimizer(self.network.parameters())
+                if self.optimizer_kwargs.get("_partial_", False)
+                else optimizer
             )
         if self.scheduler_kwargs is not None:
-            self.scheduler = utils.get_scheduler_from_kwargs(
-                self.optimizer, **self.scheduler_kwargs
+            scheduler = instantiate(self.scheduler_kwargs)
+            self.scheduler = (
+                scheduler(self.optimizer)
+                if self.scheduler_kwargs.get("_partial_", False)
+                else scheduler
             )
 
     def save_model(
@@ -236,14 +250,14 @@ def save_model(
 
         """
         model_dict = {
-            "model_kwargs": self.model_kwargs,
+            "model_kwargs": _to_plain_container(self.model_kwargs),
             "model_state_dict": self.network.state_dict(),
             "epoch": self.epoch,
             "version": self.version,
         }
 
         if self.metadata is not None:
-            model_dict["metadata"] = self.metadata
+            model_dict["metadata"] = _to_plain_container(self.metadata)
 
         if self.context is not None:
             model_dict["context"] = self.context
@@ -252,8 +266,8 @@ def save_model(
             model_dict["event_metadata"] = self.event_metadata
 
         if save_training_info:
-            model_dict["optimizer_kwargs"] = self.optimizer_kwargs
-            model_dict["scheduler_kwargs"] = self.scheduler_kwargs
+            model_dict["optimizer_kwargs"] = _to_plain_container(self.optimizer_kwargs)
+            model_dict["scheduler_kwargs"] = _to_plain_container(self.scheduler_kwargs)
             if self.optimizer is not None:
                 model_dict["optimizer_state_dict"] = self.optimizer.state_dict()
             if self.scheduler is not None:
@@ -335,7 +349,7 @@ def load_model(
 
         self.epoch = d["epoch"]
 
-        self.metadata = d["metadata"]
+        self.metadata = OmegaConf.create(d["metadata"])
 
         if "context" in d:
             self.context = d["context"]
@@ -397,7 +411,7 @@ def train(
 
         if test_only:
             test_loss = test_epoch(self, test_loader)
-            log.info(f"test loss: {test_loss:.3f}")
+            logger.info(f"test loss: {test_loss:.3f}")
 
         else:
             while not runtime_limits.limits_exceeded(self.epoch):
@@ -406,24 +420,24 @@ def train(
                 # Training
                 lr = utils.get_lr(self.optimizer)
                 with threadpool_limits(limits=1, user_api="blas"):
-                    log.info(f"\nStart training epoch {self.epoch} with lr {lr}")
+                    logger.info(f"\nStart training epoch {self.epoch} with lr {lr}")
                     time_start = time.time()
                     train_loss = train_epoch(self, train_loader)
                     train_time = time.time() - time_start
 
-                    log.info(
+                    logger.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(train_time, 60)
                         )
                     )
 
                     # Testing
-                    log.info(f"Start testing epoch {self.epoch}")
+                    logger.info(f"Start testing epoch {self.epoch}")
                     time_start = time.time()
                     test_loss = test_epoch(self, test_loader)
                     test_time = time.time() - time_start
 
-                    log.info(
+                    logger.info(
                         "Done. This took {:2.0f}:{:2.0f} min.".format(
                             *divmod(time.time() - time_start, 60)
                         )
@@ -452,7 +466,7 @@ def train(
                             }
                         )
                     except ImportError:
-                        log.warning("wandb not installed. Skipping logging to wandb.")
+                        logger.warning("wandb not installed. Skipping logging to wandb.")
 
                 if early_stopping is not None:
                     # Whether to use train or test loss
@@ -467,9 +481,9 @@ def train(
                             join(train_dir, "best_model.pt"), save_training_info=False
                         )
                     if early_stopping.early_stop:
-                        log.info("Early stopping")
+                        logger.info("Early stopping")
                         break
-                log.info(f"Finished training epoch {self.epoch}.\n")
+                logger.info(f"Finished training epoch {self.epoch}.\n")
 
 
 def train_epoch(pm, dataloader):
diff --git a/dingo/core/posterior_models/build_model.py b/dingo/core/posterior_models/build_model.py
index 2f7571032..27d79fed3 100644
--- a/dingo/core/posterior_models/build_model.py
+++ b/dingo/core/posterior_models/build_model.py
@@ -1,7 +1,5 @@
 from dingo.core.posterior_models.base_model import BasePosteriorModel
-from dingo.core.posterior_models.flow_matching import FlowMatchingPosteriorModel
-from dingo.core.posterior_models.normalizing_flow import NormalizingFlowPosteriorModel
-from dingo.core.posterior_models.score_matching import ScoreDiffusionPosteriorModel
+from hydra.utils import instantiate
 from dingo.core.utils.backward_compatibility import (
     torch_load_with_fallback,
     update_model_config,
@@ -36,12 +34,6 @@ def build_model_from_kwargs(
             "Either a filename or a settings dict must be provided, but not both."
         )
 
-    models_dict = {
-        "normalizing_flow": NormalizingFlowPosteriorModel,
-        "flow_matching": FlowMatchingPosteriorModel,
-        "score_matching": ScoreDiffusionPosteriorModel,
-    }
-
     if filename is not None:
         d, _ = torch_load_with_fallback(filename, preferred_map_location="meta")
         if "version" in d:
@@ -50,56 +42,13 @@ def build_model_from_kwargs(
             # version was introduced in v0.3.3
             check_minimum_version("dingo=0.3.2")
         update_model_config(d["metadata"]["train_settings"]["model"])  # Backward compat
-        posterior_model_type = d["metadata"]["train_settings"]["model"][
-            "posterior_model_type"
-        ]
+        model_settings = d["metadata"]["train_settings"]["model"]
     else:
         update_model_config(settings["train_settings"]["model"])  # Backward compat
-        posterior_model_type = settings["train_settings"]["model"][
-            "posterior_model_type"
-        ]
-
-    if not posterior_model_type.lower() in models_dict:
-        raise ValueError("No valid posterior model type specified.")
-
-    model = models_dict[posterior_model_type.lower()]
-
-    return model(model_filename=filename, metadata=settings, **kwargs)
-
-
-def autocomplete_model_kwargs(model_kwargs: dict, data_sample: list):
-    """
-    Autocomplete the model kwargs from train_settings and data_sample from the dataloader:
-
-    * set input dimension of embedding net to shape of data_sample[1]
-    * set dimension of parameter space to len(data_sample[0])
-    * set added_context flag of embedding net if required for gnpe proxies
-    * set context dim of posterior model to output dim of embedding net + gnpe proxy dim
+        model_settings = settings["train_settings"]["model"]
 
-    Parameters
-    ----------
-    model_kwargs: dict
-        Model settings, which are modified in-place.
-    data_sample: list
-        Sample from dataloader (e.g., wfd[0]) used for autocomplection.
-        Should be of format [parameters, GW data, gnpe_proxies], where the
-        last element is only there is GNPE proxies are required.
-    """
+    if "_target_" not in model_settings:
+        raise KeyError("Model settings need a Hydra _target_.")
 
-    # set input dims from ifo_list and domain information
-    model_kwargs["embedding_kwargs"]["input_dims"] = list(data_sample[1].shape)
-    # set dimension of parameter space of posterior model
-    model_kwargs["posterior_kwargs"]["input_dim"] = len(data_sample[0])
-    # set added_context flag of embedding net if GNPE proxies are required
-    # set context dim of nsf to output dim of embedding net + GNPE proxy dim
-    try:
-        gnpe_proxy_dim = len(data_sample[2])
-        model_kwargs["embedding_kwargs"]["added_context"] = True
-        model_kwargs["posterior_kwargs"]["context_dim"] = (
-            model_kwargs["embedding_kwargs"]["output_dim"] + gnpe_proxy_dim
-        )
-    except IndexError:
-        model_kwargs["embedding_kwargs"]["added_context"] = False
-        model_kwargs["posterior_kwargs"]["context_dim"] = model_kwargs[
-            "embedding_kwargs"
-        ]["output_dim"]
+    model_target = {"_target_": model_settings["_target_"]}
+    return instantiate(model_target, model_filename=filename, metadata=settings, **kwargs)
diff --git a/dingo/core/posterior_models/cflow_base.py b/dingo/core/posterior_models/cflow_base.py
index 2f9b09bbc..16e821420 100644
--- a/dingo/core/posterior_models/cflow_base.py
+++ b/dingo/core/posterior_models/cflow_base.py
@@ -123,7 +123,7 @@ def rhs_of_joint_ode(self, t, theta_and_div_t, *context_data, hutchinson=False):
 
     def initialize_network(self):
         model_kwargs = {
-            k: v for k, v in self.model_kwargs.items() if k != "posterior_model_type"
+            k: v for k, v in self.model_kwargs.items() if not k.startswith("_")
         }
         if self.initial_weights is not None:
             model_kwargs["initial_weights"] = self.initial_weights
diff --git a/dingo/core/posterior_models/normalizing_flow.py b/dingo/core/posterior_models/normalizing_flow.py
index 8481aeda9..67ebdb6de 100644
--- a/dingo/core/posterior_models/normalizing_flow.py
+++ b/dingo/core/posterior_models/normalizing_flow.py
@@ -39,7 +39,7 @@ def __init__(self, **kwargs):
 
     def initialize_network(self):
         model_kwargs = {
-            k: v for k, v in self.model_kwargs.items() if k != "posterior_model_type"
+            k: v for k, v in self.model_kwargs.items() if not k.startswith("_")
         }
         if self.initial_weights is not None:
             model_kwargs["initial_weights"] = self.initial_weights
diff --git a/dingo/core/result.py b/dingo/core/result.py
index a2c897a67..5edcdd318 100644
--- a/dingo/core/result.py
+++ b/dingo/core/result.py
@@ -30,7 +30,7 @@
     "log_noise_evidence",
 ]
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 def _clip_weights(weights: np.ndarray, num_clip: int) -> np.ndarray:
@@ -170,23 +170,23 @@ def reset_event(self, event_dataset):
         ):
             # This is really just for notification. Actions are only taken if the
             # event metadata differ.
-            log.warning("New event data differ from existing.")
+            logger.warning("New event data differ from existing.")
         self.context = context
 
         if self.event_metadata is not None and self.event_metadata != event_metadata:
-            log.warning("Changes")
-            log.warning("=======")
+            logger.warning("Changes")
+            logger.warning("=======")
             old_minus_new = dict(freeze(self.event_metadata) - freeze(event_metadata))
-            log.warning("Old event metadata:")
+            logger.warning("Old event metadata:")
             for k in sorted(old_minus_new):
-                log.warning(f"  {k}:  {self.event_metadata[k]}")
+                logger.warning(f"  {k}:  {self.event_metadata[k]}")
 
             new_minus_old = dict(freeze(event_metadata) - freeze(self.event_metadata))
-            log.warning("New event metadata:")
+            logger.warning("New event metadata:")
             if self.importance_sampling_metadata.get("updates") is None:
                 self.importance_sampling_metadata["updates"] = {}
             for k in sorted(new_minus_old):
-                log.warning(f"  {k}:  {event_metadata[k]}")
+                logger.warning(f"  {k}:  {event_metadata[k]}")
                 self.importance_sampling_metadata["updates"][k] = event_metadata[k]
 
             self._rebuild_domain(verbose=True)
@@ -315,12 +315,12 @@ def importance_sample(self, num_processes: int = 1, **likelihood_kwargs):
         valid_samples = np.isfinite(log_prior + delta_log_prob_target)
         theta = theta.iloc[valid_samples]
 
-        log.info(f"Calculating {len(theta)} likelihoods.")
+        logger.info(f"Calculating {len(theta)} likelihoods.")
         t0 = time.time()
         log_likelihood = self.likelihood.log_likelihood_multi(
             theta, num_processes=num_processes
         )
-        log.info(f"Done. This took {time.time() - t0:.2f} seconds.")
+        logger.info(f"Done. This took {time.time() - t0:.2f} seconds.")
 
         self.log_noise_evidence = self.likelihood.log_Zn
         self.samples["log_prior"] = log_prior
@@ -497,7 +497,7 @@ def rejection_sample(
         rng = np.random.default_rng(random_state)
         weights = self.samples["weights"].to_numpy(dtype=float)
 
-        log.info(
+        logger.info(
             f"Rejection sampling: {self.num_samples} samples, "
             f"ESS = {self.effective_sample_size:.0f} "
             f"(efficiency = {100 * self.sample_efficiency:.1f}%)"
@@ -529,7 +529,7 @@ def rejection_sample(
         unweighted = unweighted.drop(
             columns=["weights", "log_prob", "delta_log_prob_target"], errors="ignore"
         )
-        log.info(f"Produced {len(unweighted)} unweighted samples.")
+        logger.info(f"Produced {len(unweighted)} unweighted samples.")
         return unweighted
 
     def parameter_subset(self, parameters):
@@ -622,12 +622,12 @@ def print_summary(self):
         Display the number of samples, and (if importance sampling is complete) the log
         evidence and number of effective samples.
         """
-        log.info(f"Number of samples: {len(self.samples)}")
+        logger.info(f"Number of samples: {len(self.samples)}")
         if self.log_evidence is not None:
-            log.info(
+            logger.info(
                 f"Log(evidence): {self.log_evidence:.3f} +- {self.log_evidence_std:.3f}"
             )
-            log.info(
+            logger.info(
                 f"Effective samples {self.n_eff:.1f}: "
                 f"(Sample efficiency = {100 * self.sample_efficiency:.2f}%)"
             )
@@ -818,7 +818,7 @@ def plot_log_probs(self, filename="log_probs.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            log.warning("Results not importance sampled. Cannot produce log_prob plot.")
+            logger.warning("Results not importance sampled. Cannot produce log_prob plot.")
 
     def plot_weights(self, filename="weights.png"):
         """Make a scatter plot of samples weights vs log proposal."""
@@ -847,7 +847,7 @@ def plot_weights(self, filename="weights.png"):
             plt.tight_layout()
             plt.savefig(filename)
         else:
-            log.warning("Results not importance sampled. Cannot plot weights.")
+            logger.warning("Results not importance sampled. Cannot plot weights.")
 
     def get_all_injection_credible_levels(
         self, keys: list[str] = None, weighted: bool = False
@@ -1023,7 +1023,7 @@ def make_pp_plot(
 
     pvalues = []
     latex_labels = get_latex_labels(results[0].prior)
-    log.info("Key: KS-test p-value")
+    logger.info("Key: KS-test p-value")
     for ii, key in enumerate(credible_levels):
         pp = np.array(
             [
@@ -1033,7 +1033,7 @@ def make_pp_plot(
         )
         pvalue = scipy.stats.kstest(credible_levels[key], "uniform").pvalue
         pvalues.append(pvalue)
-        log.info("{}: {}".format(key, pvalue))
+        logger.info("{}: {}".format(key, pvalue))
         label = "{} ({:2.3f})".format(latex_labels[key], pvalue)
         plt.plot(x_values, pp, lines[ii], label=label, **kwargs)
 
@@ -1043,7 +1043,7 @@ def make_pp_plot(
         pvalues=pvalues,
         names=list(credible_levels.keys()),
     )
-    log.info("Combined p-value: {}".format(pvals.combined_pvalue))
+    logger.info("Combined p-value: {}".format(pvals.combined_pvalue))
 
     if title:
         ax.set_title(
diff --git a/dingo/core/samplers.py b/dingo/core/samplers.py
index 5474bc5bd..a4511ac64 100644
--- a/dingo/core/samplers.py
+++ b/dingo/core/samplers.py
@@ -22,7 +22,7 @@
 # Sampler classes are motivated by the approach of Bilby.
 #
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class Sampler(object):
@@ -159,7 +159,7 @@ def _run_sampler(
             x = [x]
         else:
             if context is not None:
-                log.warning("Unconditional model. Ignoring context.")
+                logger.warning("Unconditional model. Ignoring context.")
             x = []
 
         # For a normalizing flow, we get the log_prob for "free" when sampling,
@@ -207,7 +207,7 @@ def run_sampler(
         """
         self.samples = None
 
-        log.info(f"Running sampler to generate {num_samples} samples.")
+        logger.info(f"Running sampler to generate {num_samples} samples.")
         t0 = time.time()
         if not self.unconditional_model:
             if self.context is None:
@@ -231,7 +231,7 @@ def run_sampler(
         # correction for t_ref) and represent as DataFrame.
         self._post_process(samples)
         self.samples = pd.DataFrame(samples)
-        log.info(f"Done. This took {time.time() - t0:.1f} s.")
+        logger.info(f"Done. This took {time.time() - t0:.1f} s.")
 
     def log_prob(self, samples: pd.DataFrame | dict) -> np.ndarray:
         """
@@ -431,7 +431,7 @@ def _run_sampler(
             init_samples = self.init_sampler._run_sampler(num_samples, context)
         else:
             if self.num_iterations == 1:
-                log.warning(
+                logger.warning(
                     f"Removing initial outliers, but only carrying out "
                     f"{self.num_iterations} GNPE iteration. This risks biasing "
                     f"results."
@@ -516,7 +516,7 @@ def _run_sampler(
                 p: x["extrinsic_parameters"][p] for p in self.gnpe_proxy_parameters
             }
 
-            log.info(
+            logger.info(
                 f"it {i}.\tmin pvalue: {self.iteration_tracker.pvalue_min:.3f}"
                 f"\tproxy mean: "
                 + " ".join(f"{torch.mean(v).item():.5f}" for v in proxies.values())
diff --git a/dingo/core/utils/__init__.py b/dingo/core/utils/__init__.py
index be5b05cf8..16635122b 100644
--- a/dingo/core/utils/__init__.py
+++ b/dingo/core/utils/__init__.py
@@ -2,3 +2,4 @@
 from .trainutils import *
 from .gnpeutils import *
 from .misc import *
+from .hydra_utils import *
diff --git a/dingo/core/utils/condor_utils.py b/dingo/core/utils/condor_utils.py
index cc51de919..0ab443554 100644
--- a/dingo/core/utils/condor_utils.py
+++ b/dingo/core/utils/condor_utils.py
@@ -1,8 +1,9 @@
+import logging
 import os
 from os.path import join
 import yaml
 
-from dingo.core.utils.logging_utils import logger
+logger = logging.getLogger(__name__)
 
 
 def resubmit_condor_job(train_dir, train_settings, epoch):
diff --git a/dingo/core/utils/hydra_utils.py b/dingo/core/utils/hydra_utils.py
new file mode 100644
index 000000000..02e334ec6
--- /dev/null
+++ b/dingo/core/utils/hydra_utils.py
@@ -0,0 +1,30 @@
+from copy import deepcopy
+
+from hydra.utils import instantiate
+from omegaconf import OmegaConf
+
+
+def instantiate_with_runtime_dependencies(config, runtime_dependencies: dict):
+    """
+    Instantiate a Hydra config, optionally completing a partial with runtime objects.
+
+    If the config contains ``_runtime_dependencies_``, only those named dependencies
+    are passed to the partial. Otherwise all runtime dependencies are passed.
+    """
+    if OmegaConf.is_config(config):
+        config = OmegaConf.to_container(config, resolve=True)
+    else:
+        config = deepcopy(config)
+    dependency_names = config.pop("_runtime_dependencies_", None)
+    obj = instantiate(config)
+    if not config.get("_partial_", False):
+        return obj
+
+    if dependency_names is None:
+        dependencies = runtime_dependencies
+    else:
+        dependencies = {
+            name: runtime_dependencies[name]
+            for name in dependency_names
+        }
+    return obj(**dependencies)
diff --git a/dingo/core/utils/logging_utils.py b/dingo/core/utils/logging_utils.py
index 9efbc83db..0ab166918 100644
--- a/dingo/core/utils/logging_utils.py
+++ b/dingo/core/utils/logging_utils.py
@@ -2,6 +2,8 @@
 import os
 import sys
 
+DINGO_LOG_FORMAT = "[%(asctime)s][%(name)s][%(levelname)s] - %(message)s"
+
 
 def check_directory_exists_and_if_not_mkdir(directory, logger):
     """Checks if the given directory exists and creates it if it does not exist
@@ -18,21 +20,25 @@ def check_directory_exists_and_if_not_mkdir(directory, logger):
         logger.debug(f"Directory {directory} exists")
 
 
-def setup_logger(outdir=None, label=None, log_level="INFO"):
+def setup_logger(outdir=None, label=None, log_level="INFO", use_bilby: bool = True):
     """Setup logging output: call at the start of the script to use.
 
-    Sets up the dingo logger and, if bilby is available, also calls bilby's
-    setup_logger so that both loggers share the same file handler and their
-    entries appear together in the same log file.
+    Sets up the dingo logger and, optionally, also calls bilby's setup_logger so
+    that both loggers share the same file handler when a log file is requested.
 
     Parameters
     ----------
-    outdir, label: str
-        If supplied, write the logging output to outdir/label.log
+    outdir: str, optional
+        If supplied, write the logging output to this directory.
+    label: str, optional
+        Name of the log file without suffix. If omitted and outdir is supplied,
+        the entrypoint name is used.
     log_level: str, optional
         ['debug', 'info', 'warning']
         Either a string from the list above, or an integer as specified
         in https://docs.python.org/2/library/logging.html#logging-levels
+    use_bilby:
+        If true and bilby is available, share bilby's file handler.
     """
     if "-v" in sys.argv or "--verbose" in sys.argv:
         log_level = "DEBUG"
@@ -45,15 +51,23 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
     else:
         level = int(log_level)
 
+    formatter = logging.Formatter(DINGO_LOG_FORMAT)
+    save_log = outdir is not None
+    entrypoint = os.path.splitext(os.path.basename(sys.argv[0]))[0]
+    if not entrypoint or entrypoint.startswith("-"):
+        entrypoint = "dingo"
+    log_label = label or entrypoint
+
     # Set up bilby's logger first so we can share its file handler.
     bilby_logger = None
-    try:
-        from bilby.core.utils.log import setup_logger as bilby_setup_logger
+    if use_bilby and save_log:
+        try:
+            from bilby.core.utils.log import setup_logger as bilby_setup_logger
 
-        bilby_setup_logger(outdir=outdir or ".", label=label, log_level=log_level)
-        bilby_logger = logging.getLogger("bilby")
-    except ImportError:
-        pass
+            bilby_setup_logger(outdir=outdir, label=log_label, log_level=log_level)
+            bilby_logger = logging.getLogger("bilby")
+        except ImportError:
+            pass
 
     logger = logging.getLogger("dingo")
     logger.propagate = False
@@ -64,46 +78,28 @@ def setup_logger(outdir=None, label=None, log_level="INFO"):
         isinstance(h, logging.StreamHandler) and not isinstance(h, logging.FileHandler)
         for h in logger.handlers
     ):
-        stream_handler = logging.StreamHandler()
-        stream_handler.setFormatter(
-            logging.Formatter(
-                "%(asctime)s %(name)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
-            )
-        )
+        stream_handler = logging.StreamHandler(sys.stdout)
+        stream_handler.setFormatter(formatter)
         stream_handler.setLevel(level)
         logger.addHandler(stream_handler)
 
     # Share bilby's file handler so both loggers write to the same log file.
     if bilby_logger is not None:
         for handler in bilby_logger.handlers:
+            handler.setFormatter(formatter)
+            handler.setLevel(level)
             if isinstance(handler, logging.FileHandler):
-                handler.setFormatter(
-                    logging.Formatter(
-                        "%(asctime)s %(name)s %(levelname)-8s: %(message)s",
-                        datefmt="%H:%M",
-                    )
-                )
                 if not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
                     logger.addHandler(handler)
 
     # Fall back to a standalone file handler if bilby is unavailable.
-    if label and not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
-        if outdir:
-            check_directory_exists_and_if_not_mkdir(outdir, logger)
-        else:
-            outdir = "."
-        log_file = f"{outdir}/{label}.log"
+    if save_log and not any(isinstance(h, logging.FileHandler) for h in logger.handlers):
+        check_directory_exists_and_if_not_mkdir(outdir, logger)
+        log_file = os.path.join(outdir, f"{log_label}.log")
         file_handler = logging.FileHandler(log_file)
-        file_handler.setFormatter(
-            logging.Formatter(
-                "%(asctime)s %(name)s %(levelname)-8s: %(message)s", datefmt="%H:%M"
-            )
-        )
+        file_handler.setFormatter(formatter)
         file_handler.setLevel(level)
         logger.addHandler(file_handler)
 
     for handler in logger.handlers:
         handler.setLevel(level)
-
-
-logger = logging.getLogger("dingo")
diff --git a/dingo/core/utils/pt_to_hdf5.py b/dingo/core/utils/pt_to_hdf5.py
index 4c24c69df..f88154f91 100644
--- a/dingo/core/utils/pt_to_hdf5.py
+++ b/dingo/core/utils/pt_to_hdf5.py
@@ -1,16 +1,16 @@
 import os
 import argparse
 import logging
-import sys
 
 import torch
 import h5py
 import json
 
+from dingo.core.utils.logging_utils import setup_logger
 
-log = logging.getLogger(__name__)
-logging.captureWarnings(True)
 
+logger = logging.getLogger(__name__)
+logging.captureWarnings(True)
 
 
 def parse_args():
@@ -44,17 +44,8 @@ def parse_args():
     return parser.parse_args()
 
 
-def configure_logging():
-    logging.basicConfig(
-        level=logging.INFO,
-        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
-        stream=sys.stdout,
-        force=True,
-    )
-
-
 def main():
-    configure_logging()
+    setup_logger(use_bilby=False)
     args = parse_args()
 
     if os.path.splitext(args.in_file)[-1] != ".pt":
@@ -66,7 +57,7 @@ def main():
     # This is required for use on CVMFS
     root, ext = os.path.splitext(args.out_file)
     out_file_name = f"{root}_v{args.model_version_number}{ext}"
-    log.info(f"Output will be written to {out_file_name}")
+    logger.info(f"Output will be written to {out_file_name}")
 
     # Load data into CPU memory since we'll be saving it using CPU libraries
     d = torch.load(args.in_file, map_location=torch.device("cpu"))
diff --git a/dingo/core/utils/torchutils.py b/dingo/core/utils/torchutils.py
index 2d109d7bb..69488a2a1 100644
--- a/dingo/core/utils/torchutils.py
+++ b/dingo/core/utils/torchutils.py
@@ -5,6 +5,7 @@
 from torch.utils.data import DataLoader
 from typing import Union, Tuple, Iterable
 import bilby
+from hydra.utils import instantiate
 
 
 def fix_random_seeds(_):
@@ -84,20 +85,10 @@ def get_optimizer_from_kwargs(
     -------
     optimizer
     """
-    optimizers_dict = {
-        "adagrad": torch.optim.Adagrad,
-        "adam": torch.optim.Adam,
-        "adamw": torch.optim.AdamW,
-        "lbfgs": torch.optim.LBFGS,
-        "RMSprop": torch.optim.RMSprop,
-        "sgd": torch.optim.SGD,
-    }
-    if not "type" in optimizer_kwargs:
-        raise KeyError("Optimizer type needs to be specified.")
-    if not optimizer_kwargs["type"].lower() in optimizers_dict:
-        raise ValueError("No valid optimizer specified.")
-    optimizer = optimizers_dict[optimizer_kwargs.pop("type")]
-    return optimizer(model_parameters, **optimizer_kwargs)
+    optimizer = instantiate(optimizer_kwargs)
+    if optimizer_kwargs.get("_partial_", False):
+        return optimizer(model_parameters)
+    return optimizer
 
 
 def get_scheduler_from_kwargs(
@@ -122,17 +113,10 @@ def get_scheduler_from_kwargs(
     -------
     scheduler
     """
-    schedulers_dict = {
-        "step": torch.optim.lr_scheduler.StepLR,
-        "cosine": torch.optim.lr_scheduler.CosineAnnealingLR,
-        "reduce_on_plateau": torch.optim.lr_scheduler.ReduceLROnPlateau,
-    }
-    if not "type" in scheduler_kwargs:
-        raise KeyError("Scheduler type needs to be specified.")
-    if not scheduler_kwargs["type"].lower() in schedulers_dict:
-        raise ValueError("No valid scheduler specified.")
-    scheduler = schedulers_dict[scheduler_kwargs.pop("type")]
-    return scheduler(optimizer, **scheduler_kwargs)
+    scheduler = instantiate(scheduler_kwargs)
+    if scheduler_kwargs.get("_partial_", False):
+        return scheduler(optimizer)
+    return scheduler
 
 
 def perform_scheduler_step(
diff --git a/dingo/core/utils/trainutils.py b/dingo/core/utils/trainutils.py
index a420e8d82..4d58eb2d8 100644
--- a/dingo/core/utils/trainutils.py
+++ b/dingo/core/utils/trainutils.py
@@ -6,7 +6,7 @@
 import csv
 from typing import Literal
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class AvgTracker:
@@ -88,7 +88,7 @@ def __call__(self, val_loss: float):
         elif score < self.best_score + self.delta:
             self.counter += 1
             if self.verbose:
-                log.info(
+                logger.info(
                     f"EarlyStopping counter: {self.counter} out of {self.patience}"
                 )
             if self.counter >= self.patience:
@@ -131,7 +131,7 @@ def print_info(self, batch_idx):
         if batch_idx % self.print_freq == 0:
             td, td_avg = self.times["Dataloader"].x, self.times["Dataloader"].get_avg()
             tn, tn_avg = self.times["Network"].x, self.times["Network"].get_avg()
-            log.info(
+            logger.info(
                 "{} Epoch: {} [{}/{} ({:.0f}%)]\t\t"
                 "Loss: {:.3f} ({:.3f})\t\t"
                 "Time Dataloader: {:.3f} ({:.3f})\t\t"
@@ -203,7 +203,7 @@ def limits_exceeded(self, epoch: int = None):
         # check time limit for run
         if self.max_time_per_run is not None:
             if time.time() - self.time_start >= self.max_time_per_run:
-                log.info(
+                logger.info(
                     f"Stop run: Time limit of {self.max_time_per_run} s " f"exceeded."
                 )
                 return True
@@ -212,14 +212,14 @@ def limits_exceeded(self, epoch: int = None):
             if epoch is None:
                 raise ValueError("epoch required")
             if epoch - self.epoch_start >= self.max_epochs_per_run:
-                log.info(
+                logger.info(
                     f"Stop run: Epoch limit of {self.max_epochs_per_run} per run reached."
                 )
                 return True
         # check total epoch limit
         if self.max_epochs_total is not None:
             if epoch >= self.max_epochs_total:
-                log.info(
+                logger.info(
                     f"Stop run: Total epoch limit of {self.max_epochs_total} reached."
                 )
                 return True
@@ -324,13 +324,13 @@ def save_model(pm, log_dir, model_prefix="model", checkpoint_epochs=None):
     """
     # save current model
     model_name = join(log_dir, f"{model_prefix}_latest.pt")
-    log.info(f"Saving model to {model_name}.")
+    logger.info(f"Saving model to {model_name}.")
     pm.save_model(model_name, save_training_info=True)
-    log.info("Done.")
+    logger.info("Done.")
 
     # potentially copy model to a checkpoint
     if checkpoint_epochs is not None and pm.epoch % checkpoint_epochs == 0:
         model_name_cp = join(log_dir, f"{model_prefix}_{pm.epoch:03d}.pt")
-        log.info(f"Copy model to checkpoint {model_name_cp}.")
+        logger.info(f"Copy model to checkpoint {model_name_cp}.")
         copyfile(model_name, model_name_cp)
-        log.info("Done.")
+        logger.info("Done.")
diff --git a/dingo/gw/SVD.py b/dingo/gw/SVD.py
index 2cad5b40d..eaf003c1f 100644
--- a/dingo/gw/SVD.py
+++ b/dingo/gw/SVD.py
@@ -6,7 +6,7 @@
 from sklearn.utils.extmath import randomized_svd
 from dingo.core.dataset import DingoDataset
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class SVDBasis(DingoDataset):
@@ -160,15 +160,15 @@ def print_validation_summary(self):
                 if "mismatch" in col:
                     n = int(col.split(sep="=")[-1])
                     mismatches = self.mismatches[col]
-                    log.info(f"n = {n}")
-                    log.info("  Mean mismatch = {}".format(np.mean(mismatches)))
-                    log.info("  Standard deviation = {}".format(np.std(mismatches)))
-                    log.info("  Max mismatch = {}".format(np.max(mismatches)))
-                    log.info("  Median mismatch = {}".format(np.median(mismatches)))
-                    log.info("  Percentiles:")
-                    log.info("    99    -> {}".format(np.percentile(mismatches, 99)))
-                    log.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-                    log.info(
+                    logger.info(f"n = {n}")
+                    logger.info("  Mean mismatch = {}".format(np.mean(mismatches)))
+                    logger.info("  Standard deviation = {}".format(np.std(mismatches)))
+                    logger.info("  Max mismatch = {}".format(np.max(mismatches)))
+                    logger.info("  Median mismatch = {}".format(np.median(mismatches)))
+                    logger.info("  Percentiles:")
+                    logger.info("    99    -> {}".format(np.percentile(mismatches, 99)))
+                    logger.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
+                    logger.info(
                         "    99.99 -> {}".format(np.percentile(mismatches, 99.99))
                     )
 
diff --git a/dingo/gw/conversion/spin_conversion.py b/dingo/gw/conversion/spin_conversion.py
index f2a40b45b..2549d475f 100644
--- a/dingo/gw/conversion/spin_conversion.py
+++ b/dingo/gw/conversion/spin_conversion.py
@@ -1,5 +1,6 @@
 # Transformations between PE spins and Cartesian spins
 from functools import partial
+import logging
 from multiprocessing import Pool
 
 import numpy as np
@@ -8,7 +9,7 @@
 import lalsimulation as LS
 from threadpoolctl import threadpool_limits
 
-from dingo.core.utils.logging_utils import logger
+logger = logging.getLogger(__name__)
 
 
 DINGO_PE_SPIN_PARAMETERS = (
diff --git a/dingo/gw/data/data_download.py b/dingo/gw/data/data_download.py
index c0e2f6ed1..4b0b4d2ed 100644
--- a/dingo/gw/data/data_download.py
+++ b/dingo/gw/data/data_download.py
@@ -1,11 +1,14 @@
+import logging
+
 import numpy as np
 from gwpy.timeseries import TimeSeries
 import pycbc.psd
 import math
 
-from dingo.core.utils.logging_utils import logger
 from dingo.gw.gwutils import get_window
 
+logger = logging.getLogger(__name__)
+
 
 def download_psd(det, time_start, time_psd, window, f_s):
     """
diff --git a/dingo/gw/data/data_preparation.py b/dingo/gw/data/data_preparation.py
index 4f355b8b4..729b57881 100644
--- a/dingo/gw/data/data_preparation.py
+++ b/dingo/gw/data/data_preparation.py
@@ -1,16 +1,18 @@
 from os.path import isfile
+import logging
 
 import numpy as np
 from gwpy.timeseries import TimeSeries
 
 from dingo.core.dataset import DingoDataset
-from dingo.core.utils.logging_utils import logger
 from dingo.core.utils.misc import recursive_check_dicts_are_equal
 from dingo.gw.data.data_download import download_raw_data
 from dingo.gw.gwutils import get_window
 from dingo.gw.domains import UniformFrequencyDomain
 from dingo.gw.domains import build_domain_from_model_metadata
 
+logger = logging.getLogger(__name__)
+
 
 def load_raw_data(time_event, settings, event_dataset=None):
     """
diff --git a/dingo/gw/dataset/compression.py b/dingo/gw/dataset/compression.py
new file mode 100644
index 000000000..f45202c8b
--- /dev/null
+++ b/dingo/gw/dataset/compression.py
@@ -0,0 +1,94 @@
+import copy
+import logging
+from functools import partial
+from typing import Dict
+
+import numpy as np
+import pandas as pd
+from bilby.gw.prior import BBHPriorDict
+from torchvision.transforms import Compose
+
+from dingo.core.utils.misc import call_func_strict_output_dim
+from dingo.gw.SVD import SVDBasis
+from dingo.gw.dataset.waveform_dataset import WaveformDataset
+from dingo.gw.waveform_generator import WaveformGenerator
+
+logger = logging.getLogger(__name__)
+
+
+def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int) -> SVDBasis:
+    """
+    Train (and optionally validate) an SVD basis.
+    """
+    train_data = np.vstack([val[:n_train] for val in dataset.polarizations.values()])
+    test_data = np.vstack([val[n_train:] for val in dataset.polarizations.values()])
+    test_parameters = pd.concat(
+        [
+            # I would like to save the polarization, but saving the dataframe with
+            # string columns causes problems. Fix this later.
+            # dataset.parameters.iloc[n_train:].assign(polarization=pol)
+            dataset.parameters.iloc[n_train:]
+            for pol in dataset.polarizations
+        ]
+    )
+    test_parameters.reset_index(drop=True, inplace=True)
+
+    logger.info("Building SVD basis.")
+    basis = SVDBasis()
+    basis.generate_basis(train_data, size)
+
+    assert np.allclose(basis.V[: dataset.domain.min_idx], 0)
+
+    if test_data.size != 0:
+        basis.compute_test_mismatches(
+            test_data, parameters=test_parameters, verbose=True
+        )
+
+    return basis
+
+
+def load_svd_basis(file_name: str) -> SVDBasis:
+    """Load an existing SVD basis for Hydra-configured compression."""
+    return SVDBasis(file_name=file_name)
+
+
+def train_svd_basis_from_waveforms(
+    size: int,
+    num_training_samples: int,
+    num_validation_samples: int = 0,
+    *,
+    waveform_generator: WaveformGenerator,
+    prior: BBHPriorDict,
+    num_processes: int,
+    settings: Dict,
+    compression_transforms: list,
+    compression_settings: list,
+) -> SVDBasis:
+    """Train an SVD basis from waveforms generated with preceding compression steps."""
+    from dingo.gw.dataset.generate_dataset import generate_parameters_and_polarizations
+
+    waveform_generator.transform = Compose(compression_transforms)
+
+    num_samples = num_training_samples + num_validation_samples
+    func = partial(
+        generate_parameters_and_polarizations,
+        waveform_generator,
+        prior,
+        num_processes=num_processes,
+    )
+    parameters, polarizations = call_func_strict_output_dim(func, num_samples)
+
+    svd_dataset_settings = copy.deepcopy(settings)
+    svd_dataset_settings["num_samples"] = len(parameters)
+    svd_dataset_settings["compression"] = copy.deepcopy(compression_settings) or None
+
+    # We build a WaveformDataset containing the SVD-training waveforms because when
+    # constructed, it automatically zeroes waveforms below f_min.
+    svd_dataset = WaveformDataset(
+        dictionary={
+            "parameters": parameters,
+            "polarizations": polarizations,
+            "settings": svd_dataset_settings,
+        }
+    )
+    return train_svd_basis(svd_dataset, size, num_training_samples)
diff --git a/dingo/gw/dataset/evaluate_multibanded_domain.py b/dingo/gw/dataset/evaluate_multibanded_domain.py
index 07a27d635..10bf7a44d 100644
--- a/dingo/gw/dataset/evaluate_multibanded_domain.py
+++ b/dingo/gw/dataset/evaluate_multibanded_domain.py
@@ -3,20 +3,16 @@
 
 import hydra
 import numpy as np
+from hydra.utils import instantiate
 from omegaconf import DictConfig, OmegaConf
 from scipy.interpolate import interp1d
 
 from dingo.gw.dataset import generate_parameters_and_polarizations
-from dingo.gw.domains import build_domain, MultibandedFrequencyDomain
+from dingo.gw.domains import MultibandedFrequencyDomain
 from dingo.gw.gwutils import get_mismatch
-from dingo.gw.prior import build_prior_with_defaults
-from dingo.gw.waveform_generator import (
-    NewInterfaceWaveformGenerator,
-    WaveformGenerator,
-    generate_waveforms_parallel,
-)
+from dingo.gw.waveform_generator import generate_waveforms_parallel
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -38,40 +34,28 @@ def evaluate_multibanding(
     #
     # (a) Set geocent_time = 0.12 s (boundary of usual prior + Earth-radius crossing time)
     # (b) Set chirp mass to bottom end of prior.
-    prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    settings["intrinsic_prior"]["geocent_time"] = 0.12
-    settings["intrinsic_prior"]["chirp_mass"] = prior["chirp_mass"].minimum
+    prior = instantiate(settings["intrinsic_prior"])
+    settings["intrinsic_prior"]["dictionary"]["geocent_time"] = 0.12
+    settings["intrinsic_prior"]["dictionary"]["chirp_mass"] = prior[
+        "chirp_mass"
+    ].minimum
     # Rebuild prior with updated settings.
-    prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    log.info("Prior")
+    prior = instantiate(settings["intrinsic_prior"])
+    logger.info("Prior")
     for k, v in prior.items():
-        log.info(f"{k}: {v}")
+        logger.info(f"{k}: {v}")
 
-    domain = build_domain(settings["domain"])
-    log.info("\nDomain")
-    log.info(domain.domain_dict)
+    domain = instantiate(settings["domain"])
+    logger.info("\nDomain")
+    logger.info(domain.domain_dict)
 
     if not isinstance(domain, MultibandedFrequencyDomain):
         raise ValueError("Waveform dataset domain not a MultibandedFrequencyDomain.")
 
-    if settings["waveform_generator"].get("new_interface", False):
-        waveform_generator_mfd = NewInterfaceWaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
-        waveform_generator_ufd = NewInterfaceWaveformGenerator(
-            domain=domain.base_domain,
-            **settings["waveform_generator"],
-        )
-    else:
-        waveform_generator_mfd = WaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
-        waveform_generator_ufd = WaveformGenerator(
-            domain=domain.base_domain,
-            **settings["waveform_generator"],
-        )
+    waveform_generator_mfd = instantiate(settings["waveform_generator"], domain=domain)
+    waveform_generator_ufd = instantiate(
+        settings["waveform_generator"], domain=domain.base_domain
+    )
 
     # Generate MFD waveforms.
     parameters, polarizations_mfd = generate_parameters_and_polarizations(
@@ -96,23 +80,23 @@ def evaluate_multibanding(
                 asd_file="aLIGO_ZERO_DET_high_P_asd.txt",
             )
 
-    log.info(
+    logger.info(
         "\nMismatches between UFD waveforms and MFD waveforms interpolated to MFD."
     )
-    log.info(
+    logger.info(
         "This is a conservative estimate of the MFD performance when training "
         "networks."
     )
     mismatches = np.concatenate([v for v in mismatches.values()])
-    log.info(f"num_samples = {num_samples}")
-    log.info("  Mean mismatch = {}".format(np.mean(mismatches)))
-    log.info("  Standard deviation = {}".format(np.std(mismatches)))
-    log.info("  Max mismatch = {}".format(np.max(mismatches)))
-    log.info("  Median mismatch = {}".format(np.median(mismatches)))
-    log.info("  Percentiles:")
-    log.info("    99    -> {}".format(np.percentile(mismatches, 99)))
-    log.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
-    log.info("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
+    logger.info(f"num_samples = {num_samples}")
+    logger.info("  Mean mismatch = {}".format(np.mean(mismatches)))
+    logger.info("  Standard deviation = {}".format(np.std(mismatches)))
+    logger.info("  Max mismatch = {}".format(np.max(mismatches)))
+    logger.info("  Median mismatch = {}".format(np.median(mismatches)))
+    logger.info("  Percentiles:")
+    logger.info("    99    -> {}".format(np.percentile(mismatches, 99)))
+    logger.info("    99.9  -> {}".format(np.percentile(mismatches, 99.9)))
+    logger.info("    99.99 -> {}".format(np.percentile(mismatches, 99.99)))
 
 
 @hydra.main(
diff --git a/dingo/gw/dataset/generate_dataset.py b/dingo/gw/dataset/generate_dataset.py
index 39c525d69..c4ac0f386 100644
--- a/dingo/gw/dataset/generate_dataset.py
+++ b/dingo/gw/dataset/generate_dataset.py
@@ -9,23 +9,21 @@
 import numpy as np
 import pandas as pd
 from bilby.gw.prior import BBHPriorDict
+from hydra.utils import instantiate
 from omegaconf import DictConfig, OmegaConf
 from threadpoolctl import threadpool_limits
 from torchvision.transforms import Compose
 
+from dingo.core.utils.hydra_utils import instantiate_with_runtime_dependencies
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
-from dingo.gw.domains import build_domain
-from dingo.gw.prior import build_prior_with_defaults
-from dingo.gw.SVD import ApplySVD, SVDBasis
-from dingo.gw.transforms import WhitenFixedASD
+from dingo.gw.SVD import ApplySVD
 from dingo.gw.waveform_generator import (
-    NewInterfaceWaveformGenerator,
     WaveformGenerator,
     generate_waveforms_parallel,
 )
 from dingo.core.utils.misc import call_func_strict_output_dim
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -50,7 +48,7 @@ def generate_parameters_and_polarizations(
     pandas DataFrame of parameters
     dictionary of numpy arrays corresponding to waveform polarizations
     """
-    log.info("Generating dataset of size " + str(num_samples))
+    logger.info("Generating dataset of size " + str(num_samples))
     parameters = pd.DataFrame(prior.sample(num_samples))
 
     if num_processes > 1:
@@ -70,12 +68,12 @@ def generate_parameters_and_polarizations(
         polarizations_ok = {k: v[idx_ok] for k, v in polarizations.items()}
         parameters_ok = parameters.iloc[idx_ok]
         failed_percent = 100 * len(idx_failed) / len(parameters)
-        log.warning(
+        logger.warning(
             f"{len(idx_failed)} out of {len(parameters)} configuration ({failed_percent:.1f}%) failed to generate."
         )
         with pd.option_context("display.max_rows", None, "display.max_columns", None):
-            log.warning(parameters.iloc[idx_failed].to_string())
-        log.warning(
+            logger.warning(parameters.iloc[idx_failed].to_string())
+        logger.warning(
             f"Only returning the {len(idx_ok)} successfully generated configurations."
         )
         return parameters_ok, polarizations_ok
@@ -83,63 +81,20 @@ def generate_parameters_and_polarizations(
     return parameters, polarizations
 
 
-def train_svd_basis(dataset: WaveformDataset, size: int, n_train: int):
-    """
-    Train (and optionally validate) an SVD basis.
-
-    Parameters
-    ----------
-    dataset : WaveformDataset
-        Contains waveforms to be used for building SVD.
-    size : int
-        Number of elements to keep for the SVD basis.
-    n_train : int
-        Number of training waveforms to use. Remaining are used for validation. Note
-        that the actual number of training waveforms is n_train * len(polarizations),
-        since there is one waveform used for each polarization.
-
-    Returns
-    -------
-    SVDBasis, n_train, n_test
-        Since EOB waveforms can fail to generate, provide also the number used in
-        training and validation.
-    """
-    # Prepare data for training and validation.
-    train_data = np.vstack([val[:n_train] for val in dataset.polarizations.values()])
-    test_data = np.vstack([val[n_train:] for val in dataset.polarizations.values()])
-    test_parameters = pd.concat(
-        [
-            # I would like to save the polarization, but saving the dataframe with
-            # string columns causes problems. Fix this later.
-            # dataset.parameters.iloc[n_train:].assign(polarization=pol)
-            dataset.parameters.iloc[n_train:]
-            for pol in dataset.polarizations
-        ]
-    )
-    test_parameters.reset_index(drop=True, inplace=True)
-
-    log.info("Building SVD basis.")
-    basis = SVDBasis()
-    basis.generate_basis(train_data, size)
-
-    assert np.allclose(basis.V[: dataset.domain.min_idx], 0)
-
-    # Since there is a possibility that the size of the dataset returned by
-    # generate_parameters_and_polarizations is smaller than requested, we don't assume
-    # that there are n_test samples. Instead we just look at the size of the test
-    # dataset.
-    if test_data.size != 0:
-        basis.compute_test_mismatches(
-            test_data, parameters=test_parameters, verbose=True
+def _instantiate_compression_transform(transform_config: Dict, runtime_dependencies: Dict):
+    transform_config = copy.deepcopy(transform_config)
+    if "svd_basis" not in transform_config:
+        return instantiate_with_runtime_dependencies(
+            transform_config,
+            runtime_dependencies,
         )
 
-    # Return also the true number of samples. Some EOB waveforms may have failed to
-    # generate, so this could be smaller than the number requested.
-    n_ifos = len(dataset.polarizations)
-    n_train = len(train_data) // n_ifos
-    n_test = len(test_data) // n_ifos
-
-    return basis, n_train, n_test
+    svd_basis_config = transform_config.pop("svd_basis")
+    svd_basis = instantiate_with_runtime_dependencies(
+        svd_basis_config,
+        runtime_dependencies,
+    )
+    return instantiate(transform_config, svd_basis=svd_basis)
 
 
 def _settings_from_config(cfg: DictConfig) -> Dict:
@@ -164,83 +119,35 @@ def generate_dataset(settings: Dict, num_processes: int) -> WaveformDataset:
     A WaveformDataset based on the settings.
     """
 
-    prior = build_prior_with_defaults(settings["intrinsic_prior"])
-    domain = build_domain(settings["domain"])
-
-    new_interface_flag = settings["waveform_generator"].get("new_interface", False)
-    if new_interface_flag:
-        waveform_generator = NewInterfaceWaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
-    else:
-        waveform_generator = WaveformGenerator(
-            domain=domain,
-            **settings["waveform_generator"],
-        )
+    prior = instantiate(settings["intrinsic_prior"])
+    domain = instantiate(settings["domain"])
+    waveform_generator = instantiate(settings["waveform_generator"], domain=domain)
 
     dataset_dict = {"settings": settings}
 
     if settings.get("compression", None) is not None:
         compression_transforms = []
-
-        if "whitening" in settings["compression"]:
-            compression_transforms.append(
-                WhitenFixedASD(
-                    domain,
-                    asd_file=settings["compression"]["whitening"],
-                    inverse=False,
-                )
+        compression_settings = []
+        runtime_dependencies = {
+            "domain": domain,
+            "waveform_generator": waveform_generator,
+            "prior": prior,
+            "num_processes": num_processes,
+            "settings": settings,
+            "compression_transforms": compression_transforms,
+            "compression_settings": compression_settings,
+        }
+
+        for transform_config in settings["compression"]:
+            compression_transform = _instantiate_compression_transform(
+                transform_config,
+                runtime_dependencies,
             )
+            compression_transforms.append(compression_transform)
+            compression_settings.append(copy.deepcopy(transform_config))
 
-        if "svd" in settings["compression"]:
-            svd_settings = settings["compression"]["svd"]
-
-            # Load an SVD basis from file, if specified.
-            if "file" in svd_settings:
-                basis = SVDBasis(file_name=svd_settings["file"])
-
-            # Otherwise, generate the basis based on simulated waveforms.
-            else:
-                # If using whitened waveforms, then the SVD should be based on these.
-                waveform_generator.transform = Compose(compression_transforms)
-
-                n_train = svd_settings["num_training_samples"]
-                n_test = svd_settings.get("num_validation_samples", 0)
-
-                func = partial(
-                    generate_parameters_and_polarizations,
-                    waveform_generator,
-                    prior,
-                    num_processes=num_processes,
-                )
-                parameters, polarizations = call_func_strict_output_dim(
-                    func, n_train + n_test
-                )
-                svd_dataset_settings = copy.deepcopy(settings)
-                svd_dataset_settings["num_samples"] = len(parameters)
-                del svd_dataset_settings["compression"]["svd"]
-
-                # We build a WaveformDataset containing the SVD-training waveforms
-                # because when constructed, it will automatically zero the waveforms
-                # below f_min. This is useful for EOB waveforms, which are Fourier
-                # transformed from time domain, and hence are nonzero below f_min. The
-                # waveforms need to be zeroed below f_min because this corresponds to
-                # setting the lower bound of the likelihood integral.
-
-                svd_dataset = WaveformDataset(
-                    dictionary={
-                        "parameters": parameters,
-                        "polarizations": polarizations,
-                        "settings": svd_dataset_settings,
-                    }
-                )
-                basis, n_train, n_test = train_svd_basis(
-                    svd_dataset, svd_settings["size"], n_train
-                )
-
-            compression_transforms.append(ApplySVD(basis))
-            dataset_dict["svd"] = basis.to_dictionary()
+            if isinstance(compression_transform, ApplySVD):
+                dataset_dict["svd"] = compression_transform.svd_basis.to_dictionary()
 
         waveform_generator.transform = Compose(compression_transforms)
 
diff --git a/dingo/gw/dataset/generate_dataset_dag.py b/dingo/gw/dataset/generate_dataset_dag.py
index 718149f72..d6ab967ba 100644
--- a/dingo/gw/dataset/generate_dataset_dag.py
+++ b/dingo/gw/dataset/generate_dataset_dag.py
@@ -1,11 +1,14 @@
 import os
 import argparse
+import logging
 from typing import Dict
 from pycondor import Job, Dagman
 import yaml
 import copy
 
-from dingo.core.utils.logging_utils import logger, setup_logger
+from dingo.core.utils.logging_utils import setup_logger
+
+logger = logging.getLogger(__name__)
 
 # Fixed file names
 svd_fn = "svd.hdf5"
diff --git a/dingo/gw/dataset/utils.py b/dingo/gw/dataset/utils.py
index d0b8e8cc9..73440e153 100644
--- a/dingo/gw/dataset/utils.py
+++ b/dingo/gw/dataset/utils.py
@@ -1,7 +1,6 @@
 import copy
 import logging
 import argparse
-import sys
 import textwrap
 from typing import List
 
@@ -9,23 +8,14 @@
 import numpy as np
 import yaml
 
-from dingo.gw.SVD import SVDBasis
-from dingo.gw.dataset.generate_dataset import train_svd_basis
+from dingo.core.utils.logging_utils import setup_logger
+from dingo.gw.dataset.compression import train_svd_basis
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
-def configure_logging():
-    logging.basicConfig(
-        level=logging.INFO,
-        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
-        stream=sys.stdout,
-        force=True,
-    )
-
-
 def merge_datasets(dataset_list: List[WaveformDataset]) -> WaveformDataset:
     """
     Merge a collection of datasets into one.
@@ -41,7 +31,7 @@ def merge_datasets(dataset_list: List[WaveformDataset]) -> WaveformDataset:
     WaveformDataset containing the merged data.
     """
 
-    log.info(f"Merging {len(dataset_list)} datasets into one.")
+    logger.info(f"Merging {len(dataset_list)} datasets into one.")
 
     # This ensures that all the keys are copied into the new dataset. The "extensive"
     # parts of the dataset (parameters, waveforms) will be overwritten by the combined
@@ -74,7 +64,6 @@ def merge_datasets_cli():
     parallelized waveform generation.
     """
 
-    configure_logging()
     parser = argparse.ArgumentParser(
         formatter_class=argparse.RawDescriptionHelpFormatter,
         description=textwrap.dedent(
@@ -103,8 +92,7 @@ def merge_datasets_cli():
         "--settings_file", type=str, help="YAML file containing new dataset settings."
     )
     args = parser.parse_args()
-    out_path = Path(args.out_file)
-    setup_logger(outdir=str(out_path.parent), label=out_path.stem)
+    setup_logger(use_bilby=False)
 
     dataset_list = []
     for i in range(args.num_parts):
@@ -121,7 +109,7 @@ def merge_datasets_cli():
         merged_dataset.settings = settings
 
     merged_dataset.to_file(args.out_file)
-    log.info(
+    logger.info(
         f"Complete. New dataset consists of {merged_dataset.settings['num_samples']} "
         f"samples."
     )
@@ -132,7 +120,6 @@ def build_svd_cli():
     Command-line function to build an SVD based on an uncompressed dataset file.
     """
 
-    configure_logging()
     parser = argparse.ArgumentParser(
         formatter_class=argparse.RawDescriptionHelpFormatter,
         description=textwrap.dedent(
@@ -162,8 +149,7 @@ def build_svd_cli():
         ),
     )
     args = parser.parse_args()
-    out_path = Path(args.out_file)
-    setup_logger(outdir=str(out_path.parent), label=out_path.stem)
+    setup_logger(use_bilby=False)
 
     dataset = WaveformDataset(file_name=args.dataset_file)
     if args.num_train is None:
@@ -174,7 +160,7 @@ def build_svd_cli():
     basis, n_train, n_test = train_svd_basis(dataset, args.size, n_train)
     # FIXME: This is not an ideal treatment. We should update the waveform generation
     #  to always provide the requested number of waveforms.
-    log.info(
+    logger.info(
         f"SVD basis trained based on {n_train} waveforms and validated on {n_test} "
         f"waveforms. Note that if this differs from number requested, it will not be "
         f"reflected in the settings file. This is likely due to EOB failure to "
diff --git a/dingo/gw/dataset/waveform_dataset.py b/dingo/gw/dataset/waveform_dataset.py
index 43d298c5e..35c996d00 100644
--- a/dingo/gw/dataset/waveform_dataset.py
+++ b/dingo/gw/dataset/waveform_dataset.py
@@ -3,12 +3,12 @@
 import numpy as np
 import torch.utils.data
 from hydra.utils import instantiate
+from omegaconf import OmegaConf
 from torchvision.transforms import Compose
 
 from dingo.core.dataset import DingoDataset, recursive_hdf5_load
 from dingo.gw.SVD import SVDBasis, ApplySVD
 from dingo.gw.domains import build_domain
-from dingo.gw.transforms import WhitenFixedASD
 
 
 class WaveformDataset(DingoDataset, torch.utils.data.Dataset):
@@ -155,15 +155,36 @@ def update_domain(self, domain_update: Optional[dict] = None):
         # Determine where any domain adjustment must be applied. If the dataset is SVD
         # compressed, then adjust the SVD matrices. Otherwise, adjust the dataset
         # itself (if it already has been loaded).
-        if (
-            self.settings.get("compression", None) is not None
-            and "svd" in self.settings["compression"]
-        ):
+        if self._compression_has_svd():
             self.svd["V"] = self.domain.update_data(self.svd["V"], axis=0)
         elif self.polarizations is not None:
             for k, v in self.polarizations.items():
                 self.polarizations[k] = self.domain.update_data(v)
 
+    def _compression_entries(self):
+        compression = self.settings.get("compression", None)
+        if compression is None:
+            return []
+        if isinstance(compression, dict) or OmegaConf.is_dict(compression):
+            entries = []
+            if "whitening" in compression:
+                entries.append(
+                    {
+                        "_target_": "dingo.gw.transforms.WhitenFixedASD",
+                        "asd_file": compression["whitening"],
+                    }
+                )
+            if "svd" in compression:
+                entries.append({"_target_": "dingo.gw.SVD.ApplySVD"})
+            return entries
+        return list(compression)
+
+    def _compression_has_svd(self):
+        return any(
+            entry.get("_target_", "").endswith("ApplySVD")
+            for entry in self._compression_entries()
+        )
+
     def initialize_decompression(self, svd_size_update: Optional[int] = None):
         """
         Sets up decompression transforms. These are applied to the raw dataset before
@@ -179,37 +200,47 @@ def initialize_decompression(self, svd_size_update: Optional[int] = None):
         # These transforms must be in reverse order compared to when dataset was
         # constructed.
 
-        if "svd" in self.settings["compression"]:
-            assert self.svd is not None
-
-            # We allow the option to reduce the size of the SVD used for decompression,
-            # since decompression is the costliest preprocessing operation. Be careful
-            # when using this to not introduce a large mismatch.
-            if svd_size_update is not None:
-                if svd_size_update > self.svd["V"].shape[-1] or svd_size_update < 0:
-                    raise ValueError(
-                        f"Cannot truncate SVD from size "
-                        f"{self.svd['V'].shape[-1]} to size "
-                        f"{svd_size_update}."
+        for compression_entry in reversed(self._compression_entries()):
+            target = compression_entry.get("_target_", "")
+
+            if target.endswith("ApplySVD"):
+                assert self.svd is not None
+
+                # We allow the option to reduce the size of the SVD used for
+                # decompression, since decompression is the costliest preprocessing
+                # operation. Be careful when using this to not introduce a large
+                # mismatch.
+                if svd_size_update is not None:
+                    if svd_size_update > self.svd["V"].shape[-1] or svd_size_update < 0:
+                        raise ValueError(
+                            f"Cannot truncate SVD from size "
+                            f"{self.svd['V'].shape[-1]} to size "
+                            f"{svd_size_update}."
+                        )
+                    self.svd["V"] = self.svd["V"][:, :svd_size_update]
+                    self.svd["s"] = self.svd["s"][:svd_size_update]
+                    if self.polarizations is not None:
+                        for k, v in self.polarizations.items():
+                            self.polarizations[k] = v[:, :svd_size_update]
+
+                svd_basis = SVDBasis(dictionary=self.svd)
+                decompression_transform_list.append(ApplySVD(svd_basis, inverse=True))
+
+            elif target.endswith("WhitenFixedASD"):
+                transform_config = OmegaConf.to_container(
+                    OmegaConf.create(compression_entry),
+                    resolve=True,
+                )
+                transform_config.pop("_partial_", None)
+                transform_config.pop("_runtime_dependencies_", None)
+                decompression_transform_list.append(
+                    instantiate(
+                        transform_config,
+                        domain=self.domain,
+                        inverse=True,
+                        precision=self.precision,
                     )
-                self.svd["V"] = self.svd["V"][:, :svd_size_update]
-                self.svd["s"] = self.svd["s"][:svd_size_update]
-                if self.polarizations is not None:
-                    for k, v in self.polarizations.items():
-                        self.polarizations[k] = v[:, :svd_size_update]
-
-            svd_basis = SVDBasis(dictionary=self.svd)
-            decompression_transform_list.append(ApplySVD(svd_basis, inverse=True))
-
-        if "whitening" in self.settings["compression"]:
-            decompression_transform_list.append(
-                WhitenFixedASD(
-                    self.domain,
-                    asd_file=self.settings["compression"]["whitening"],
-                    inverse=True,
-                    precision=self.precision,
                 )
-            )
 
         self.decompression_transform = Compose(decompression_transform_list)
 
diff --git a/dingo/gw/domains/build_domain.py b/dingo/gw/domains/build_domain.py
index b0820d87d..63bd4c40a 100644
--- a/dingo/gw/domains/build_domain.py
+++ b/dingo/gw/domains/build_domain.py
@@ -4,6 +4,7 @@
     MultibandedFrequencyDomain,
     TimeDomain,
 )
+from hydra.utils import instantiate
 
 
 def build_domain(settings: dict) -> Domain:
@@ -20,10 +21,13 @@ def build_domain(settings: dict) -> Domain:
     -------
     A Domain instance of the correct type.
     """
+    if "_target_" in settings:
+        return instantiate(settings)
+
     if "type" not in settings:
         raise ValueError(
-            f'Domain settings must include a "type" key. Settings included '
-            f"the keys {settings.keys()}."
+            'Domain settings must include a "_target_" or "type" key. '
+            f"Settings included the keys {settings.keys()}."
         )
 
     # The settings other than 'type' correspond to the kwargs of the Domain constructor.
@@ -60,8 +64,13 @@ def build_domain_from_model_metadata(
     A Domain instance of the correct type.
     """
     domain = build_domain(model_metadata["dataset_settings"]["domain"])
-    if "domain_update" in model_metadata["train_settings"]["data"]:
-        domain.update(model_metadata["train_settings"]["data"]["domain_update"])
+    data_settings = model_metadata["train_settings"]["data"]
+    domain_update = data_settings.get("waveform_dataset", {}).get(
+        "domain_update",
+        data_settings.get("domain_update"),
+    )
+    if domain_update is not None:
+        domain.update(domain_update)
     if base and hasattr(domain, "base_domain"):
         domain = domain.base_domain
     return domain
diff --git a/dingo/gw/download_strain_data.py b/dingo/gw/download_strain_data.py
index 87aa55693..a76a7b528 100644
--- a/dingo/gw/download_strain_data.py
+++ b/dingo/gw/download_strain_data.py
@@ -1,13 +1,16 @@
+import logging
+
 import numpy as np
 import pycbc.psd
 from gwpy.timeseries import TimeSeries
 
-from dingo.core.utils.logging_utils import logger
 from dingo.gw.domains import UniformFrequencyDomain
 from dingo.gw.gwutils import (
     get_window,
 )
 
+logger = logging.getLogger(__name__)
+
 
 def estimate_single_psd(
     time_start,
diff --git a/dingo/gw/gwutils.py b/dingo/gw/gwutils.py
index f2b0730c0..cce4eca8d 100644
--- a/dingo/gw/gwutils.py
+++ b/dingo/gw/gwutils.py
@@ -6,7 +6,6 @@
 
 from bilby.gw.detector import PowerSpectralDensity
 
-from dingo.gw.prior import default_extrinsic_dict
 from dingo.gw.prior import BBHExtrinsicPriorDict
 
 
@@ -26,18 +25,6 @@ def get_window(window_kwargs):
         raise NotImplementedError(f"Unknown window type {type}.")
 
 
-def get_extrinsic_prior_dict(extrinsic_prior):
-    """Build dict for extrinsic prior by starting with
-    default_extrinsic_dict, and overwriting every element for which
-    extrinsic_prior is not default.
-    TODO: Move to dingo.gw.prior.py?"""
-    extrinsic_prior_dict = default_extrinsic_dict.copy()
-    for k, v in extrinsic_prior.items():
-        if v.lower() != "default":
-            extrinsic_prior_dict[k] = v
-    return extrinsic_prior_dict
-
-
 def get_mismatch(a, b, domain, asd_file=None):
     """
     Mistmatch is 1 - overlap, where overlap is defined by
@@ -100,7 +87,10 @@ def get_standardization_dict(
     # Some of the extrinsic prior parameters have analytic means and standard
     # deviations. If possible, this will either get these, or else it will estimate
     # them numerically.
-    ext_prior = BBHExtrinsicPriorDict(extrinsic_prior_dict)
+    if isinstance(extrinsic_prior_dict, BBHExtrinsicPriorDict):
+        ext_prior = extrinsic_prior_dict
+    else:
+        ext_prior = BBHExtrinsicPriorDict(extrinsic_prior_dict)
     mean_extrinsic, std_extrinsic = ext_prior.mean_std(ext_prior.keys())
 
     # Check that overlap between intrinsic and extrinsic parameters is only
diff --git a/dingo/gw/importance_sampling/diagnostics.py b/dingo/gw/importance_sampling/diagnostics.py
index 1733dbcbf..ffa0b489c 100644
--- a/dingo/gw/importance_sampling/diagnostics.py
+++ b/dingo/gw/importance_sampling/diagnostics.py
@@ -7,7 +7,7 @@
 from dingo.core.utils.plotting import plot_corner_multi
 from dingo.gw.result import Result
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 def plot_posterior_slice2d(
@@ -222,7 +222,7 @@ def plot_diagnostics(
     inds = np.where(weights > threshold)[0]
     theta_new = theta.loc[inds]
     weights_new = weights[inds]
-    log.info(
+    logger.info(
         f"Generating cornerplot with {len(theta_new)} out of {len(theta)} IS samples."
     )
 
diff --git a/dingo/gw/importance_sampling/importance_weights.py b/dingo/gw/importance_sampling/importance_weights.py
index 94e922868..83ddcb9d0 100644
--- a/dingo/gw/importance_sampling/importance_weights.py
+++ b/dingo/gw/importance_sampling/importance_weights.py
@@ -18,7 +18,7 @@
 from dingo.gw.inference.gw_samplers import GWSampler
 from dingo.gw.importance_sampling.diagnostics import plot_diagnostics
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -108,20 +108,20 @@ def main(cfg: DictConfig):
         event_name = str(result.event_metadata["time_event"])
         nde_name = settings["nde"].get("path") or join(outdir, f"nde-{event_name}.pt")
         if isfile(nde_name):
-            log.info(f"Loading nde at {nde_name} for event {event_name}.")
+            logger.info(f"Loading nde at {nde_name} for event {event_name}.")
             nde = NormalizingFlowPosteriorModel(
                 model_filename=nde_name,
                 device=settings["nde"]["training"]["device"],
                 load_training_info=False,
             )
         else:
-            log.info(f"Training new nde for event {event_name}.")
+            logger.info(f"Training new nde for event {event_name}.")
             nde = result.train_unconditional_flow(
                 inference_parameters,
                 settings["nde"],
                 train_dir=outdir,
             )
-            log.info(f"Renaming trained nde model to {nde_name}.")
+            logger.info(f"Renaming trained nde model to {nde_name}.")
             rename(join(outdir, "model_latest.pt"), nde_name)
 
         # Step 1a: Sample from proposal.
@@ -170,10 +170,10 @@ def main(cfg: DictConfig):
     # print(np.std(log_evidences) / np.mean(log_evidences_std))
 
     if synthetic_phase:
-        log.info("Sampling synthetic phase.")
+        logger.info("Sampling synthetic phase.")
         result.sample_synthetic_phase(synthetic_phase_kwargs)
 
-    log.info("Importance sampling.")
+    logger.info("Importance sampling.")
     result.importance_sample(
         num_processes=settings.get("num_processes", 1),
         time_marginalization_kwargs=time_marginalization_kwargs,
@@ -187,7 +187,7 @@ def main(cfg: DictConfig):
     diagnostics_dir = join(outdir, "IS-diagnostics")
     if not exists(diagnostics_dir):
         makedirs(diagnostics_dir)
-    log.info("Plotting diagnostics.")
+    logger.info("Plotting diagnostics.")
     plot_diagnostics(
         result,
         diagnostics_dir,
diff --git a/dingo/gw/inference/gw_samplers.py b/dingo/gw/inference/gw_samplers.py
index ecfabf822..1774bc23f 100644
--- a/dingo/gw/inference/gw_samplers.py
+++ b/dingo/gw/inference/gw_samplers.py
@@ -1,4 +1,5 @@
 import logging
+from copy import deepcopy
 from typing import Union, Protocol
 
 import numpy as np
@@ -6,26 +7,24 @@
 from astropy.time import Time
 from bilby.core.prior import PriorDict, DeltaFunction, Constraint
 from bilby.gw.detector import InterferometerList
+from hydra.utils import instantiate
 from torchvision.transforms import Compose
 
 from dingo.core.samplers import Sampler, GNPESampler
 from dingo.core.transforms import GetItem, RenameKey
+from dingo.core.utils.hydra_utils import instantiate_with_runtime_dependencies
 from dingo.gw.domains import (
     MultibandedFrequencyDomain,
     build_domain_from_model_metadata,
     UniformFrequencyDomain,
     Domain,
 )
-from dingo.gw.domains import build_domain
-from dingo.gw.gwutils import get_extrinsic_prior_dict
-from dingo.gw.prior import build_prior_with_defaults
 from dingo.gw.result import Result
 from dingo.gw.transforms import (
     WhitenAndScaleStrain,
     RepackageStrainsAndASDS,
     ToTorch,
     SelectStandardizeRepackageParameters,
-    GNPECoalescenceTimes,
     TimeShiftStrain,
     GNPEBase,
     PostCorrectGeocentTime,
@@ -35,7 +34,7 @@
     MaskDataForFrequencyRangeUpdate,
 )
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class SamplerProtocol(Protocol):
@@ -167,13 +166,12 @@ def _build_domain(self: Sampler):
 
         Called by __init__() immediately after _build_prior().
         """
-        self.domain = build_domain(
-            self.base_model_metadata["dataset_settings"]["domain"]
-        )
+        self.domain = instantiate(self.base_model_metadata["dataset_settings"]["domain"])
 
         data_settings = self.base_model_metadata["train_settings"]["data"]
-        if "domain_update" in data_settings:
-            self.domain.update(data_settings["domain_update"])
+        domain_update = data_settings.get("waveform_dataset", {}).get("domain_update")
+        if domain_update is not None:
+            self.domain.update(domain_update)
 
     def _correct_reference_time(
         self: Sampler, samples: Union[dict, pd.DataFrame], inverse: bool = False
@@ -229,11 +227,11 @@ def _post_process(self, samples: Union[dict, pd.DataFrame], inverse: bool = Fals
             Whether to apply instead the inverse transformation. This is used prior to
             calculating the log_prob.
         """
-        intrinsic_prior = self.metadata["dataset_settings"]["intrinsic_prior"]
-        extrinsic_prior = get_extrinsic_prior_dict(
+        prior = instantiate(self.metadata["dataset_settings"]["intrinsic_prior"])
+        extrinsic_prior = instantiate(
             self.metadata["train_settings"]["data"]["extrinsic_prior"]
         )
-        prior = build_prior_with_defaults({**intrinsic_prior, **extrinsic_prior})
+        prior.update(extrinsic_prior)
 
         if not inverse:
             # Add fixed parameters from prior.
@@ -241,7 +239,7 @@ def _post_process(self, samples: Union[dict, pd.DataFrame], inverse: bool = Fals
             for k, p in prior.items():
                 if isinstance(p, DeltaFunction) and k not in samples:
                     v = p.peak
-                    log.info(f"Adding fixed parameter {k} = {v} from prior.")
+                    logger.info(f"Adding fixed parameter {k} = {v} from prior.")
                     samples[k] = p.peak * np.ones(num_samples)
         else:
             # Drop non-inference parameters from samples.
@@ -446,12 +444,12 @@ def _initialize_transforms(self):
         transform_pre = []
         transform_pre.append(RenameKey("data", "waveform"))
         if gnpe_time_settings:
+            gnpe_time_settings = deepcopy(gnpe_time_settings)
+            gnpe_time_settings["inference"] = True
             transform_pre.append(
-                GNPECoalescenceTimes(
-                    ifo_list,
-                    gnpe_time_settings["kernel"],
-                    gnpe_time_settings["exact_equiv"],
-                    inference=True,
+                instantiate_with_runtime_dependencies(
+                    gnpe_time_settings,
+                    {"ifo_list": ifo_list},
                 )
             )
             transform_pre.append(TimeShiftStrain(ifo_list, self.domain))
@@ -473,8 +471,8 @@ def _initialize_transforms(self):
                 self.gnpe_parameters += transform.input_parameter_names
                 for k, v in transform.kernel.items():
                     self.gnpe_kernel[k] = v
-        log.info(f"GNPE parameters: {self.gnpe_parameters}")
-        log.info(f"GNPE kernel: {self.gnpe_kernel}")
+        logger.info(f"GNPE parameters: {self.gnpe_parameters}")
+        logger.info(f"GNPE kernel: {self.gnpe_kernel}")
 
         self.transform_pre = Compose(transform_pre)
 
diff --git a/dingo/gw/injection.py b/dingo/gw/injection.py
index f1679fb47..352a7417c 100644
--- a/dingo/gw/injection.py
+++ b/dingo/gw/injection.py
@@ -1,16 +1,17 @@
+import logging
+
 import numpy as np
 from bilby.gw.detector import InterferometerList
+from hydra.utils import instantiate
 from torchvision.transforms import Compose
 
-from dingo.core.utils.logging_utils import logger
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.domains import (
     UniformFrequencyDomain,
     MultibandedFrequencyDomain,
 )
-from dingo.gw.domains import build_domain, build_domain_from_model_metadata
-from dingo.gw.gwutils import get_extrinsic_prior_dict
-from dingo.gw.prior import build_prior_with_defaults, split_off_extrinsic_parameters
+from dingo.gw.domains import build_domain_from_model_metadata
+from dingo.gw.prior import split_off_extrinsic_parameters
 from dingo.gw.transforms import (
     GetDetectorTimes,
     ProjectOntoDetectors,
@@ -23,6 +24,8 @@
     NewInterfaceWaveformGenerator,
 )
 
+logger = logging.getLogger(__name__)
+
 
 class GWSignal(object):
     """
@@ -355,16 +358,16 @@ def from_posterior_model_metadata(cls, metadata):
         metadata : dict
             Dict which you can get via PosteriorModel.metadata
         """
-        intrinsic_prior = metadata["dataset_settings"]["intrinsic_prior"]
-        extrinsic_prior = get_extrinsic_prior_dict(
+        prior = instantiate(metadata["dataset_settings"]["intrinsic_prior"])
+        extrinsic_prior = instantiate(
             metadata["train_settings"]["data"]["extrinsic_prior"]
         )
-        prior = build_prior_with_defaults({**intrinsic_prior, **extrinsic_prior})
+        prior.update(extrinsic_prior)
 
         return cls(
             prior=prior,
             wfg_kwargs=metadata["dataset_settings"]["waveform_generator"],
-            wfg_domain=build_domain(metadata["dataset_settings"]["domain"]),
+            wfg_domain=instantiate(metadata["dataset_settings"]["domain"]),
             data_domain=build_domain_from_model_metadata(metadata),
             ifo_list=metadata["train_settings"]["data"]["detectors"],
             t_ref=metadata["train_settings"]["data"]["ref_time"],
diff --git a/dingo/gw/likelihood.py b/dingo/gw/likelihood.py
index f48893ad2..6336de53e 100644
--- a/dingo/gw/likelihood.py
+++ b/dingo/gw/likelihood.py
@@ -1,11 +1,13 @@
 from multiprocessing import Pool
 from typing import Optional
+import logging
 
 import numpy as np
 import pandas as pd
 from scipy.fft import fft
 from scipy.special import logsumexp
 from bilby.gw.utils import ln_i0
+from hydra.utils import instantiate
 from threadpoolctl import threadpool_limits
 
 from dingo.core.likelihood import Likelihood
@@ -22,7 +24,8 @@
 )
 from dingo.gw.domains import build_domain
 from dingo.gw.data.data_preparation import get_event_data_and_domain
-from dingo.core.utils.logging_utils import logger
+
+logger = logging.getLogger(__name__)
 
 
 class StationaryGaussianGWLikelihood(GWSignal, Likelihood):
@@ -590,7 +593,7 @@ def _log_likelihood_time_marginalized(self, theta):
             kappa2_ij[:, j] = np.sum(kappa2_, axis=0)
         # Marginalize over time; this requires multiplying the likelihoods with the
         # prior (*not* in log space), summing over the time bins (both axes i and j!),
-        # and then taking the log. See Eq. (52) in https://arxiv.org/pdf/1809.02293.pdf.
+        # and then taking the logger. See Eq. (52) in https://arxiv.org/pdf/1809.02293.pdf.
         # To prevent numerical issues, we use the logsumexp trick.
         assert kappa2_ij.shape == self.time_prior_log.shape
         exponent = kappa2_ij + self.time_prior_log
@@ -791,7 +794,7 @@ def build_stationary_gaussian_likelihood(
     # set up likelihood
     likelihood = StationaryGaussianGWLikelihood(
         wfg_kwargs=metadata["model"]["dataset_settings"]["waveform_generator"],
-        wfg_domain=build_domain(metadata["model"]["dataset_settings"]["domain"]),
+        wfg_domain=instantiate(metadata["model"]["dataset_settings"]["domain"]),
         data_domain=data_domain,
         event_data=event_data,
         t_ref=metadata["event"]["time_event"],
diff --git a/dingo/gw/ls_cli.py b/dingo/gw/ls_cli.py
index 6abd3d72e..7db853b02 100644
--- a/dingo/gw/ls_cli.py
+++ b/dingo/gw/ls_cli.py
@@ -1,7 +1,6 @@
 import argparse
 import json
 import logging
-import sys
 from pathlib import Path
 from pprint import pformat
 
@@ -11,12 +10,13 @@
 from dingo.core.dataset import DingoDataset
 from dingo.core.result import Result
 from dingo.core.utils.backward_compatibility import torch_load_with_fallback
+from dingo.core.utils.logging_utils import setup_logger
 from dingo.gw.dataset import WaveformDataset
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.SVD import SVDBasis
 
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -27,23 +27,18 @@ def parse_args():
 
 
 def ls():
-    logging.basicConfig(
-        level=logging.INFO,
-        format="[%(asctime)s][%(name)s][%(levelname)s] - %(message)s",
-        stream=sys.stdout,
-        force=True,
-    )
+    setup_logger(use_bilby=False)
     args = parse_args()
     file_name = args.file_name
 
     path = Path(file_name)
     if path.suffix == ".pt":
-        log.info("Extracting information about torch model.\n")
+        logger.info("Extracting information about torch model.\n")
         d, _ = torch_load_with_fallback(path, preferred_map_location="meta")
-        log.info(f"Version: {d.get('version')}\n")
-        log.info(f"Model epoch: {d['epoch']}\n")
-        log.info("Model metadata:")
-        log.info(
+        logger.info(f"Version: {d.get('version')}\n")
+        logger.info(f"Model epoch: {d['epoch']}\n")
+        logger.info("Model metadata:")
+        logger.info(
             yaml.dump(
                 d["metadata"],
                 default_flow_style=False,
@@ -57,10 +52,10 @@ def ls():
 
         if dataset_type == "gw_result" or dataset_type == "core_result":
             result = Result(file_name=file_name)
-            log.info(f"Version: {result.version}")
-            log.info("\nDingo Result\n" + "============\n")
+            logger.info(f"Version: {result.version}")
+            logger.info("\nDingo Result\n" + "============\n")
 
-            log.info(
+            logger.info(
                 "Metadata\n"
                 + "--------\n"
                 + yaml.dump(
@@ -70,7 +65,7 @@ def ls():
                 )
             )
             if result.event_metadata:
-                log.info(
+                logger.info(
                     "Event information:\n"
                     + "------------------\n"
                     + yaml.dump(
@@ -80,7 +75,7 @@ def ls():
                     ),
                 )
             if result.importance_sampling_metadata is not None:
-                log.info(
+                logger.info(
                     "Importance sampling:\n"
                     + "--------------------\n"
                     + yaml.dump(
@@ -90,28 +85,28 @@ def ls():
                     ),
                 )
             if result.log_evidence:
-                log.info("Summary:\n" + "--------")
+                logger.info("Summary:\n" + "--------")
                 result.print_summary()
 
         elif dataset_type == "svd_basis":
             svd = SVDBasis(file_name=file_name)
-            log.info(f"Dingo version: {svd.version}")
-            log.info("\nSVD Basis\n" + "=========\n")
+            logger.info(f"Dingo version: {svd.version}")
+            logger.info("\nSVD Basis\n" + "=========\n")
 
-            log.info(f"Basis size: {svd.n}.")
-            log.info("\nValidation summary:\n" + "-------------------")
+            logger.info(f"Basis size: {svd.n}.")
+            logger.info("\nValidation summary:\n" + "-------------------")
             svd.print_validation_summary()
 
         elif dataset_type == "waveform_dataset":
             waveform_dataset = WaveformDataset(
                 file_name=file_name, leave_waveforms_on_disk=True
             )
-            log.info(f"Dingo version: {waveform_dataset.version}")
-            log.info("\nWaveform dataset\n" + "================\n")
+            logger.info(f"Dingo version: {waveform_dataset.version}")
+            logger.info("\nWaveform dataset\n" + "================\n")
 
-            log.info(f"Dataset size: {len(waveform_dataset)}")
+            logger.info(f"Dataset size: {len(waveform_dataset)}")
 
-            log.info(
+            logger.info(
                 "\nSettings\n"
                 + "--------\n"
                 + yaml.dump(
@@ -123,18 +118,18 @@ def ls():
 
             if waveform_dataset.svd:
                 svd = SVDBasis(dictionary=waveform_dataset.svd)
-                log.info("\nSVD validation summary:\n" + "---------------------------")
+                logger.info("\nSVD validation summary:\n" + "---------------------------")
                 svd.print_validation_summary()
 
         elif dataset_type == "asd_dataset":
             asd_dataset = ASDDataset(file_name=file_name)
-            log.info(f"Dingo version: {asd_dataset.version}")
-            log.info("\nASD dataset\n" + "================\n")
+            logger.info(f"Dingo version: {asd_dataset.version}")
+            logger.info("\nASD dataset\n" + "================\n")
 
-            log.info(f"Dataset size: {asd_dataset.length_info}\n")
-            log.info(f"GPS times (min/max): {asd_dataset.gps_info}")
+            logger.info(f"Dataset size: {asd_dataset.length_info}\n")
+            logger.info(f"GPS times (min/max): {asd_dataset.gps_info}")
 
-            log.info(
+            logger.info(
                 "\nSettings\n"
                 + "--------\n"
                 + yaml.dump(
@@ -146,22 +141,22 @@ def ls():
 
         elif dataset_type == "trained_model":
             with h5py.File(file_name, "r") as f:
-                log.info("Extracting information about torch model.\n")
-                log.info(f"Version: {f.attrs['version']}")
-                log.info(f"Model epoch: {f.attrs['epoch']}")
-                log.info("Model metadata:")
+                logger.info("Extracting information about torch model.\n")
+                logger.info(f"Version: {f.attrs['version']}")
+                logger.info(f"Model epoch: {f.attrs['epoch']}")
+                logger.info("Model metadata:")
 
                 for d in ["model_kwargs", "metadata"]:
                     json_data = json.loads(f["serialized_dicts"][d][()])
-                    log.info(f"\n{d}:\n" + "-" * (len(d) + 1))
-                    log.info(pformat(json_data))
+                    logger.info(f"\n{d}:\n" + "-" * (len(d) + 1))
+                    logger.info(pformat(json_data))
 
         else:
             # Legacy (before dataset_type identifier).
             try:
                 svd = SVDBasis(file_name=file_name)
-                log.info(f"SVD dataset of size n={svd.n}.")
-                log.info("Validation summary:")
+                logger.info(f"SVD dataset of size n={svd.n}.")
+                logger.info("Validation summary:")
                 svd.print_validation_summary()
 
             except KeyError:
@@ -173,7 +168,7 @@ def ls():
                     ],
                 )
                 if dataset.settings is not None:
-                    log.info(
+                    logger.info(
                         yaml.dump(
                             dataset.settings,
                             default_flow_style=False,
@@ -182,13 +177,13 @@ def ls():
                     )
                 if dataset.svd is not None:
                     svd = SVDBasis(dictionary=dataset.svd)
-                    log.info("SVD validation summary:")
+                    logger.info("SVD validation summary:")
                     svd.print_validation_summary()
 
     elif path.suffix == ".yaml":
         with open(path, "r") as f:
             settings = yaml.safe_load(f)
-        log.info(
+        logger.info(
             yaml.dump(
                 settings,
                 default_flow_style=False,
@@ -197,7 +192,7 @@ def ls():
         )
 
     else:
-        log.info("File type unrecognized.")
+        logger.info("File type unrecognized.")
 
 
 def determine_dataset_type(file_name):
diff --git a/dingo/gw/noise/asd_dataset.py b/dingo/gw/noise/asd_dataset.py
index b4a378f75..6a9c57615 100644
--- a/dingo/gw/noise/asd_dataset.py
+++ b/dingo/gw/noise/asd_dataset.py
@@ -10,7 +10,7 @@
 from dingo.gw.gwutils import *
 from dingo.gw.dataset import DingoDataset
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 HIGH_ASD_VALUE = 1.0
 
@@ -62,7 +62,7 @@ def __init__(
                     self.gps_times.pop(ifo)
 
         if "window_factor" in self.settings["domain_dict"]:
-            log.info(
+            logger.info(
                 "Warning: 'window_factor' is no longer used in ASDDataset. "
                 "Removing from settings."
             )
@@ -138,14 +138,14 @@ def update_domain(self, domain_update):
             self.domain.domain_dict["type"] == "UniformFrequencyDomain"
             and domain_update["type"] == "MultibandedFrequencyDomain"
         ):
-            log.info("Updating ASD dataset to MultibandedFrequencyDomain.")
+            logger.info("Updating ASD dataset to MultibandedFrequencyDomain.")
             asd_dataset_decimated = {}
             mfd = build_domain(domain_update)
             ufd = mfd.base_domain
             if not check_domain_compatibility(self.asds, ufd):
                 # If the ASD length is not compatible with the new base UFD,
                 # first truncate it.
-                log.info(
+                logger.info(
                     f"  Truncating first to new base UniformFrequencyDomain: f_max "
                     f"{self.domain.f_max} Hz -> {ufd.f_max} Hz"
                 )
diff --git a/dingo/gw/noise/asd_estimation.py b/dingo/gw/noise/asd_estimation.py
index bc6e9e290..960483643 100644
--- a/dingo/gw/noise/asd_estimation.py
+++ b/dingo/gw/noise/asd_estimation.py
@@ -15,7 +15,7 @@
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.utils import psd_data_path
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -75,7 +75,7 @@ def download_and_estimate_psds(
     w = get_window(window_kwargs)
     asd_filename_list = {det: [] for det in detectors}
     total_segments = sum(len(time_segments[det]) for det in detectors)
-    log.info(
+    logger.info(
         f"Estimating PSDs for {total_segments} time segments across "
         f"{len(time_segments)} detector(s)."
     )
@@ -93,17 +93,17 @@ def download_and_estimate_psds(
         else:
             estimation_kwargs["det"] = det
 
-        log.info(f"Processing {len(time_segments[det])} PSD segment(s) for {det}.")
+        logger.info(f"Processing {len(time_segments[det])} PSD segment(s) for {det}.")
         for index, (start, end) in enumerate(
             tqdm(time_segments[det], disable=not verbose)
         ):
             filename = join(psd_path, f"asd_{start}.hdf5")
             asd_filename_list[det].append(filename)
             if os.path.exists(filename):
-                log.info(f"ASD file already exists, skipping {filename}.")
+                logger.info(f"ASD file already exists, skipping {filename}.")
                 continue
 
-            log.info(f"Estimating ASD for {det} segment starting at {start}.")
+            logger.info(f"Estimating ASD for {det} segment starting at {start}.")
             dataset_dict = {
                 "settings": {
                     "dataset_settings": settings["dataset_settings"],
@@ -120,7 +120,7 @@ def download_and_estimate_psds(
             dataset = ASDDataset(dictionary=dataset_dict)
             dataset.to_file(file_name=filename)
 
-    log.info("PSD estimation complete.")
+    logger.info("PSD estimation complete.")
     return asd_filename_list
 
 
diff --git a/dingo/gw/noise/generate_dataset.py b/dingo/gw/noise/generate_dataset.py
index 526a3bd94..9f7cde6d3 100644
--- a/dingo/gw/noise/generate_dataset.py
+++ b/dingo/gw/noise/generate_dataset.py
@@ -13,7 +13,7 @@
 from dingo.gw.noise.asd_dataset import ASDDataset
 from dingo.gw.noise.utils import merge_datasets, get_time_segments
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -63,11 +63,11 @@ def generate_dataset(cfg: DictConfig):
         #     pass
         #
         # dagman.build()
-        # log.info("DAG submission file written.")
+        # logger.info("DAG submission file written.")
 
     else:
 
-        log.info("Downloading strain data and estimating PSDs...")
+        logger.info("Downloading strain data and estimating PSDs...")
         asd_filename_list = download_and_estimate_psds(
             data_dir, settings, time_segments, verbose=verbose
         )
@@ -75,7 +75,7 @@ def generate_dataset(cfg: DictConfig):
             det: [ASDDataset(asd_file) for asd_file in asd_file_list]
             for det, asd_file_list in asd_filename_list.items()
         }
-        log.info("Merging single dataset files into one...")
+        logger.info("Merging single dataset files into one...")
         dataset = merge_datasets(asd_dataset_list)
         filename = out_name
         if filename is None:
diff --git a/dingo/gw/noise/synthetic/asd_sampling.py b/dingo/gw/noise/synthetic/asd_sampling.py
index 1c2f941cb..3d0ff1421 100644
--- a/dingo/gw/noise/synthetic/asd_sampling.py
+++ b/dingo/gw/noise/synthetic/asd_sampling.py
@@ -9,7 +9,7 @@
     get_index_for_elem,
 )
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class KDE:
@@ -57,7 +57,7 @@ def fit(self, weights=None):
                         weights=weights,
                     )
                 except np.linalg.LinAlgError:
-                    log.info(
+                    logger.info(
                         "Warning: Singular Matrix encountered in spectral KDE. Adding small Gaussian noise..."
                     )
                     perturbed_features = spectral_features[:, i, :] + np.random.normal(
diff --git a/dingo/gw/noise/synthetic/generate_dataset.py b/dingo/gw/noise/synthetic/generate_dataset.py
index a3896a384..b52d4bb2b 100644
--- a/dingo/gw/noise/synthetic/generate_dataset.py
+++ b/dingo/gw/noise/synthetic/generate_dataset.py
@@ -12,7 +12,7 @@
 from dingo.gw.noise.synthetic.asd_sampling import KDE, get_rescaling_params
 from dingo.gw.noise.synthetic.utils import reconstruct_psds_from_parameters
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
diff --git a/dingo/gw/noise/utils.py b/dingo/gw/noise/utils.py
index dc94d5b44..6bd9e333d 100644
--- a/dingo/gw/noise/utils.py
+++ b/dingo/gw/noise/utils.py
@@ -16,7 +16,7 @@
 from gwpy.table import EventTable
 from dingo.gw.noise.asd_dataset import ASDDataset
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 """
@@ -161,7 +161,7 @@ def merge_datasets(asd_dataset_list):
     merged_dict = {"asds": {}, "gps_times": {}}
 
     for det, asd_list in asd_dataset_list.items():
-        log.info(f"Merging {len(asd_list)} datasets into one for detector {det}.")
+        logger.info(f"Merging {len(asd_list)} datasets into one for detector {det}.")
         merged_dict["asds"][det] = np.vstack(
             [asd_dataset.asds[det] for asd_dataset in asd_list]
         )
diff --git a/dingo/gw/prior.py b/dingo/gw/prior.py
index 5fc97129e..02f47b849 100644
--- a/dingo/gw/prior.py
+++ b/dingo/gw/prior.py
@@ -1,5 +1,3 @@
-from copy import deepcopy
-
 from bilby.gw.prior import BBHPriorDict
 from bilby.gw.conversion import (
     fill_from_fixed_priors,
@@ -8,7 +6,7 @@
 from bilby.core.prior import Uniform, Sine, Cosine
 
 import numpy as np
-from typing import Dict, Set, Any
+from typing import Set, Any
 import warnings
 
 # Silence INFO and WARNING messages from bilby
@@ -99,76 +97,6 @@ def mean_std(self, keys=([]), sample_size=50000, force_numerical=False):
         return mean, std
 
 
-default_extrinsic_dict = {
-    "dec": "bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')",
-    "ra": 'bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")',
-    "geocent_time": "bilby.core.prior.Uniform(minimum=-0.1, maximum=0.1, name='geocent_time')",
-    "psi": 'bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")',
-    "luminosity_distance": "bilby.core.prior.Uniform(minimum=100.0, maximum=6000.0, name='luminosity_distance')",
-}
-
-default_intrinsic_dict = {
-    "mass_1": "bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')",
-    "mass_2": "bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')",
-    "mass_ratio": "bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')",
-    "chirp_mass": "bilby.gw.prior.UniformInComponentsChirpMass(minimum=25.0, maximum=100.0, name='chirp_mass')",
-    "luminosity_distance": 1000.0,
-    "theta_jn": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')",
-    "phase": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")',
-    "a_1": "bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')",
-    "a_2": "bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')",
-    "tilt_1": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')",
-    "tilt_2": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')",
-    "phi_12": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")',
-    "phi_jl": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")',
-    "geocent_time": 0.0,
-}
-
-default_inference_parameters = [
-    "chirp_mass",
-    "mass_ratio",
-    "phase",
-    "a_1",
-    "a_2",
-    "tilt_1",
-    "tilt_2",
-    "phi_12",
-    "phi_jl",
-    "theta_jn",
-    "luminosity_distance",
-    "geocent_time",
-    "ra",
-    "dec",
-    "psi",
-]
-
-
-def build_prior_with_defaults(prior_settings: Dict[str, str]):
-    """
-    Generate BBHPriorDict based on dictionary of prior settings,
-    allowing for default values.
-
-    Parameters
-    ----------
-    prior_settings: Dict
-        A dictionary containing prior definitions for intrinsic parameters
-        Allowed values for each parameter are:
-            * 'default' to use a default prior
-            * a string for a custom prior, e.g.,
-               "Uniform(minimum=10.0, maximum=80.0, name=None, latex_label=None, unit=None, boundary=None)"
-
-    Depending on the particular prior choices the dimensionality of a
-    parameter sample obtained from the returned GWPriorDict will vary.
-    """
-
-    full_prior_settings = deepcopy(prior_settings)
-    for k, v in prior_settings.items():
-        if v == "default":
-            full_prior_settings[k] = default_intrinsic_dict[k]
-
-    return BBHPriorDict(full_prior_settings)
-
-
 def split_off_extrinsic_parameters(theta):
     """
     Split theta into intrinsic and extrinsic parameters.
diff --git a/dingo/gw/result.py b/dingo/gw/result.py
index b5d130ba3..70b68650c 100644
--- a/dingo/gw/result.py
+++ b/dingo/gw/result.py
@@ -8,6 +8,7 @@
 from bilby.core.prior import Uniform, Constraint, PriorDict, DeltaFunction
 from bilby.gw.prior import CalibrationPriorDict
 from bilby_pipe.utils import CALIBRATION_CORRECTION_TYPE_LOOKUP
+from hydra.utils import instantiate
 
 from dingo.core.density import (
     interpolated_sample_and_log_prob_multi,
@@ -18,14 +19,12 @@
 from dingo.gw.conversion import change_spin_conversion_phase
 from dingo.gw.domains import MultibandedFrequencyDomain
 from dingo.gw.domains import build_domain
-from dingo.gw.gwutils import get_extrinsic_prior_dict
 from dingo.gw.likelihood import StationaryGaussianGWLikelihood
-from dingo.gw.prior import build_prior_with_defaults
 from dingo.core.utils.backward_compatibility import check_minimum_version
 
 
 RANDOM_STATE = 150914
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class Result(CoreResult):
@@ -168,12 +167,13 @@ def _build_domain(self):
 
         Called by __init__() immediately after _build_prior().
         """
-        self.domain = build_domain(self.base_metadata["dataset_settings"]["domain"])
+        self.domain = instantiate(self.base_metadata["dataset_settings"]["domain"])
         check_minimum_version(self.version, raise_exception=False)
 
         data_settings = self.base_metadata["train_settings"]["data"]
-        if "domain_update" in data_settings:
-            self.domain.update(data_settings["domain_update"])
+        domain_update = data_settings.get("waveform_dataset", {}).get("domain_update")
+        if domain_update is not None:
+            self.domain.update(domain_update)
 
     def _rebuild_domain(self, verbose=False):
         """Rebuild the domain based on settings updated for importance sampling.
@@ -201,8 +201,8 @@ def _rebuild_domain(self, verbose=False):
             )
 
             if verbose:
-                log.info("Rebuilding domain as follows:")
-                log.info(
+                logger.info("Rebuilding domain as follows:")
+                logger.info(
                     yaml.dump(
                         domain_dict,
                         default_flow_style=False,
@@ -212,15 +212,17 @@ def _rebuild_domain(self, verbose=False):
             self.domain = build_domain(domain_dict)
         else:
             if verbose:
-                log.info("No domain updates found; domain not rebuilt.")
+                logger.info("No domain updates found; domain not rebuilt.")
 
     def _build_prior(self):
         """Build the prior based on model metadata. Called by __init__()."""
-        intrinsic_prior = self.base_metadata["dataset_settings"]["intrinsic_prior"]
-        extrinsic_prior = get_extrinsic_prior_dict(
+        self.prior = instantiate(
+            self.base_metadata["dataset_settings"]["intrinsic_prior"]
+        )
+        extrinsic_prior = instantiate(
             self.base_metadata["train_settings"]["data"]["extrinsic_prior"]
         )
-        self.prior = build_prior_with_defaults({**intrinsic_prior, **extrinsic_prior})
+        self.prior.update(extrinsic_prior)
 
         prior_update = self.importance_sampling_metadata.get("prior_update")
         if prior_update is not None:
@@ -361,11 +363,11 @@ def _build_likelihood(
         if "updates" in self.importance_sampling_metadata:
             if "T" in self.importance_sampling_metadata["updates"]:
                 delta_f_new = 1 / self.importance_sampling_metadata["updates"]["T"]
-                log.info(
+                logger.info(
                     f'Updating waveform generation delta_f from {wfg_domain_dict["delta_f"]} to {delta_f_new}.'
                 )
                 wfg_domain_dict["delta_f"] = delta_f_new
-        wfg_domain = build_domain(wfg_domain_dict)
+        wfg_domain = instantiate(wfg_domain_dict)
 
         self.likelihood = StationaryGaussianGWLikelihood(
             wfg_kwargs=self.base_metadata["dataset_settings"]["waveform_generator"],
@@ -459,7 +461,7 @@ def sample_calibration_parameters(self, calibration_sampling_kwargs: dict):
 
         # Sample calibration parameters and calculate log_prob
         num_samples = len(self.samples)
-        log.info(f"Sampling calibration parameters for {num_samples} samples.")
+        logger.info(f"Sampling calibration parameters for {num_samples} samples.")
 
         delta_log_prob = np.zeros(num_samples)
 
@@ -578,7 +580,7 @@ def sample_synthetic_phase(
             self.synthetic_phase_kwargs.get("num_processes", 1), num_valid_samples // 10
         )
 
-        log.info(f"Estimating synthetic phase for {num_valid_samples} samples.")
+        logger.info(f"Estimating synthetic phase for {num_valid_samples} samples.")
         t0 = time.time()
 
         if not inverse:
@@ -667,7 +669,7 @@ def sample_synthetic_phase(
             self.samples["log_prob"] = log_prob_array
             del self.samples["phase"]
 
-        log.info(f"Done. This took {time.time() - t0:.2f} s.")
+        logger.info(f"Done. This took {time.time() - t0:.2f} s.")
 
     def get_samples_bilby_phase(self, num_processes=1):
         """
diff --git a/dingo/gw/training/train_builders.py b/dingo/gw/training/train_builders.py
index f3695015e..7baade18e 100755
--- a/dingo/gw/training/train_builders.py
+++ b/dingo/gw/training/train_builders.py
@@ -5,68 +5,25 @@
 import torch.multiprocessing
 import torchvision
 from threadpoolctl import threadpool_limits
-from bilby.gw.detector import InterferometerList
+from hydra.utils import instantiate
 
 from dingo.gw.SVD import SVDBasis
 
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
-from dingo.gw.domains import build_domain
 from dingo.gw.transforms import (
-    ProjectOntoDetectors,
-    SampleNoiseASD,
-    WhitenAndScaleStrain,
     AddWhiteNoiseComplex,
     SelectStandardizeRepackageParameters,
     RepackageStrainsAndASDS,
     UnpackDict,
-    GNPECoalescenceTimes,
-    SampleExtrinsicParameters,
-    GetDetectorTimes,
     CropMaskStrainRandom,
 )
-from dingo.gw.noise.asd_dataset import ASDDataset
-from dingo.gw.prior import default_inference_parameters
 from dingo.gw.gwutils import *
 from dingo.core.utils import *
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
-def build_dataset(
-    data_settings: dict,
-    leave_waveforms_on_disk: Optional[bool] = False,
-) -> WaveformDataset:
-    """Build a dataset based on a settings dictionary. This should contain the path of
-    a saved waveform dataset.
-
-    This function also truncates the dataset as necessary.
-
-    Parameters
-    ----------
-    data_settings : dict
-    leave_waveforms_on_disk: bool
-        If provided, the values associated with the waveforms will not be loaded into memory during initialization.
-        Instead, they will be loaded from disk when the dataset is accessed. This is useful for reducing the memory
-        load of large datasets, but can slow down data preprocessing.
-
-    Returns
-    -------
-    WaveformDataset
-    """
-
-    # Build and truncate datasets
-    domain_update = data_settings.get("domain_update", None)
-    wfd = WaveformDataset(
-        file_name=data_settings["waveform_dataset_path"],
-        precision="single",
-        domain_update=domain_update,
-        svd_size_update=data_settings.get("svd_size_update"),
-        leave_waveforms_on_disk=leave_waveforms_on_disk,
-    )
-    return wfd
-
-
-def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=None):
+def set_train_transforms(wfd, data_settings, asd_dataset_config, omit_transforms=None):
     """
     Set the transform attribute of a waveform dataset based on a settings dictionary.
     The transform takes waveform polarizations, samples random extrinsic parameters,
@@ -75,120 +32,62 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
 
     Note that the WaveformDataset is modified in-place, so this function returns nothing.
 
+    The Hydra training config distinguishes two transform lists:
+
+    * ``data.standardization_transforms`` is a short prefix pipeline used before this
+      function is called, in ``train_pipeline._populate_parameter_standardization``.
+      It reproduces the old procedural behavior where parameter standardizations were
+      computed after sampling extrinsic parameters, detector times, and optional GNPE
+      proxy times, but before projection/noise/repackaging. The computed mean/std
+      dictionary is written back into ``data.standardization`` and into the resolved
+      ``SelectStandardizeRepackageParameters`` config.
+    * ``data.transforms`` is the actual training pipeline instantiated here and assigned
+      to ``wfd.transform``.
+
+    Hydra entries such as ``ifo_list: {_target_: ..., interferometers:
+    "${data.detectors}"}`` mean that the target is called with keyword arguments, e.g.
+    ``InterferometerList(interferometers=data_settings["detectors"])`` after OmegaConf
+    interpolation has resolved ``${data.detectors}``.
+
     Parameters
     ----------
     wfd : WaveformDataset
     data_settings : dict
-    asd_dataset_path : str
-        Path corresponding to the ASD dataset used to generate noise.
+    asd_dataset_config : dict
+        Config for the ASD dataset used to generate noise.
     omit_transforms :
         List of sub-transforms to omit from the full composition.
     """
 
-    log.info("Setting train transforms.")
+    logger.info("Setting train transforms.")
     if omit_transforms is not None:
-        log.info("Omitting \n\t" + "\n\t".join([t.__name__ for t in omit_transforms]))
+        logger.info("Omitting \n\t" + "\n\t".join([t.__name__ for t in omit_transforms]))
 
     # By passing the wfd domain when instantiating the noise dataset, this ensures the
     # domains will match. In particular, it truncates the ASD dataset beyond the new
     # f_max, and sets it to 1 below f_min.
-    asd_dataset = ASDDataset(
-        asd_dataset_path,
-        ifos=data_settings["detectors"],
-        precision="single",
-        domain_update=wfd.domain.domain_dict,
-    )
+    asd_dataset = instantiate(asd_dataset_config, domain_update=wfd.domain.domain_dict)
     assert wfd.domain == asd_dataset.domain
-    domain = wfd.domain
-
-    extrinsic_prior_dict = get_extrinsic_prior_dict(data_settings["extrinsic_prior"])
-    if data_settings["inference_parameters"] == "default":
-        data_settings["inference_parameters"] = default_inference_parameters
-
-    ref_time = data_settings["ref_time"]
-    # Build detector objects
-    ifo_list = InterferometerList(data_settings["detectors"])
-
-    # Build transforms.
-    transforms = [
-        SampleExtrinsicParameters(extrinsic_prior_dict),
-        GetDetectorTimes(ifo_list, ref_time),
-    ]
-
-    extra_context_parameters = []
-    if "gnpe_time_shifts" in data_settings:
-        d = data_settings["gnpe_time_shifts"]
-        transforms.append(
-            GNPECoalescenceTimes(
-                ifo_list,
-                d["kernel"],
-                d["exact_equiv"],
-                inference=False,
-            )
-        )
-        extra_context_parameters += transforms[-1].context_parameters
-
-    # Add the GNPE context to context_parameters the first time the transforms are
-    # constructed. We do not want to overwrite the ordering of the parameters in
-    # subsequent runs.
-    if "context_parameters" not in data_settings:
-        data_settings["context_parameters"] = []
-    for p in extra_context_parameters:
-        if p not in data_settings["context_parameters"]:
-            data_settings["context_parameters"].append(p)
-
-    # If the standardization factors have already been set, use those. Otherwise,
-    # calculate them, and save them within the data settings.
-    #
-    # Standardizations are calculated at this point because the present set of
-    # transforms is sufficient for generating samples of all regression and context
-    # parameters.
-    try:
-        standardization_dict = data_settings["standardization"]
-        log.info("Using previously-calculated parameter standardizations.")
-    except KeyError:
-        log.info("Calculating new parameter standardizations.")
-        standardization_dict = get_standardization_dict(
-            extrinsic_prior_dict,
-            wfd,
-            data_settings["inference_parameters"] + data_settings["context_parameters"],
-            torchvision.transforms.Compose(transforms),
-        )
-        data_settings["standardization"] = standardization_dict
-
-    transforms.append(ProjectOntoDetectors(ifo_list, domain, ref_time))
-    transforms.append(SampleNoiseASD(asd_dataset))
-    transforms.append(WhitenAndScaleStrain(domain.noise_std))
-    # We typically add white detector noise. For debugging purposes, this can be turned
-    # off with zero_noise option in data_settings.
-    if not data_settings.get("zero_noise", False):
-        transforms.append(AddWhiteNoiseComplex())
-    transforms.append(
-        SelectStandardizeRepackageParameters(
-            {
-                k: data_settings[k]
-                for k in ["inference_parameters", "context_parameters"]
-            },
-            standardization_dict,
-        )
-    )
-    transforms.append(
-        RepackageStrainsAndASDS(data_settings["detectors"], first_index=domain.min_idx)
-    )
-    if "random_strain_cropping" in data_settings:
-        transforms.append(
-            CropMaskStrainRandom(domain, **data_settings["random_strain_cropping"])
-        )
-    if data_settings["context_parameters"]:
-        selected_keys = ["inference_parameters", "waveform", "context_parameters"]
-    else:
-        selected_keys = ["inference_parameters", "waveform"]
 
-    transforms.append(UnpackDict(selected_keys=selected_keys))
+    runtime_dependencies = {
+        "domain": wfd.domain,
+        "asd_dataset": asd_dataset,
+    }
+
+    transforms = []
+    for transform_config in data_settings["transforms"]:
+        transform = instantiate_with_runtime_dependencies(
+            transform_config,
+            runtime_dependencies,
+        )
+        transforms.append(transform)
 
     # Drop transforms that are not desired. This is useful for generating, e.g.,
     # noise-free data, or for producing data not formatted for input to the network.
-    if omit_transforms is not None:
+    omit_transforms = list(omit_transforms or [])
+    if data_settings.get("zero_noise", False):
+        omit_transforms.append(AddWhiteNoiseComplex)
+    if omit_transforms:
         transforms = [t for t in transforms if type(t) not in omit_transforms]
 
     wfd.transform = torchvision.transforms.Compose(transforms)
@@ -197,7 +96,7 @@ def set_train_transforms(wfd, data_settings, asd_dataset_path, omit_transforms=N
 def build_svd_for_embedding_network(
     wfd: WaveformDataset,
     data_settings: dict,
-    asd_dataset_path: str,
+    asd_dataset_config: dict,
     size: int,
     num_training_samples: int,
     num_validation_samples: int,
@@ -216,7 +115,7 @@ def build_svd_for_embedding_network(
     ----------
     wfd : WaveformDataset
     data_settings : dict
-    asd_dataset_path : str
+    asd_dataset_config : dict
         Training waveforms will be whitened with respect to these ASDs.
     size : int
         Number of basis elements to include in the SVD projection.
@@ -244,14 +143,14 @@ def build_svd_for_embedding_network(
     torch.multiprocessing.set_sharing_strategy("file_system")
 
     # Fix the luminosity distance to a standard value, just in order to generate the SVD.
-    data_settings["extrinsic_prior"]["luminosity_distance"] = "100.0"
+    data_settings["extrinsic_prior"]["dictionary"]["luminosity_distance"] = "100.0"
 
     # Build the dataset, but with certain transforms omitted. In particular, we want to
     # build the SVD based on zero-noise waveforms. They should still be whitened though.
     set_train_transforms(
         wfd,
         data_settings,
-        asd_dataset_path,
+        asd_dataset_config,
         omit_transforms=[
             AddWhiteNoiseComplex,
             RepackageStrainsAndASDS,
@@ -261,7 +160,7 @@ def build_svd_for_embedding_network(
         ],
     )
 
-    log.info("Generating waveforms for embedding network SVD initialization.")
+    logger.info("Generating waveforms for embedding network SVD initialization.")
     time_start = time.time()
     ifos = list(wfd[0]["waveform"].keys())
     waveform_len = len(wfd[0]["waveform"][ifos[0]])
@@ -299,25 +198,25 @@ def build_svd_for_embedding_network(
                 waveforms[ifo][lower : lower + n] = strains[:n]
             if lower + n == num_waveforms:
                 break
-    log.info(f"...done. This took {time.time() - time_start:.0f} s.")
+    logger.info(f"...done. This took {time.time() - time_start:.0f} s.")
 
     # Reset the standard sharing strategy.
     torch.multiprocessing.set_sharing_strategy(old_sharing_strategy)
 
-    log.info("Generating SVD basis for ifo:")
+    logger.info("Generating SVD basis for ifo:")
     time_start = time.time()
     basis_dict = {}
     for ifo in ifos:
         basis = SVDBasis()
         basis.generate_basis(waveforms[ifo][:num_training_samples], size)
         basis_dict[ifo] = basis
-        log.info(f"...{ifo} done.")
-    log.info(f"...this took {time.time() - time_start:.0f} s.")
+        logger.info(f"...{ifo} done.")
+    logger.info(f"...this took {time.time() - time_start:.0f} s.")
 
     if out_dir is not None:
-        log.info("Testing SVD basis matrices.")
+        logger.info("Testing SVD basis matrices.")
         for ifo, basis in basis_dict.items():
-            log.info(f"...{ifo}:")
+            logger.info(f"...{ifo}:")
             basis.compute_test_mismatches(
                 waveforms[ifo][num_training_samples:],
                 parameters=parameters.iloc[num_training_samples:].reset_index(
@@ -326,19 +225,19 @@ def build_svd_for_embedding_network(
                 verbose=True,
             )
             basis.to_file(os.path.join(out_dir, f"svd_{ifo}.hdf5"))
-    log.info("Done")
+    logger.info("Done")
 
     # Return V matrices in standard order. Drop the elements below domain.min_idx,
     # since the neural network expects data truncated below these. The dropped elements
     # should be 0.
-    log.info(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
-    log.info("...V matrix shapes:")
+    logger.info(f"Truncating SVD matrices below index {wfd.domain.min_idx}.")
+    logger.info("...V matrix shapes:")
     V_rb_list = []
     for ifo in data_settings["detectors"]:
         V = basis_dict[ifo].V
         assert np.allclose(V[: wfd.domain.min_idx], 0)
         V = V[wfd.domain.min_idx :]
-        log.info("      " + str(V.shape))
+        logger.info("      " + str(V.shape))
         V_rb_list.append(V)
-    log.info("\n")
+    logger.info("\n")
     return V_rb_list
diff --git a/dingo/gw/training/train_pipeline.py b/dingo/gw/training/train_pipeline.py
index abf745181..b0e4f5b03 100644
--- a/dingo/gw/training/train_pipeline.py
+++ b/dingo/gw/training/train_pipeline.py
@@ -4,24 +4,23 @@
 
 import hydra
 import numpy as np
+import torchvision
 import yaml
 import shutil
 import time
 from copy import deepcopy
 
-from hydra.utils import to_absolute_path
+from hydra.utils import instantiate, to_absolute_path
 from omegaconf import DictConfig, OmegaConf
 from threadpoolctl import threadpool_limits
 
-from dingo.core.posterior_models.build_model import (
-    autocomplete_model_kwargs,
-    build_model_from_kwargs,
-)
+from dingo.core.posterior_models.build_model import build_model_from_kwargs
+from dingo.core.utils.hydra_utils import instantiate_with_runtime_dependencies
 from dingo.gw.training.train_builders import (
-    build_dataset,
     set_train_transforms,
     build_svd_for_embedding_network,
 )
+from dingo.gw.gwutils import get_standardization_dict
 from dingo.core.utils.trainutils import RuntimeLimits
 from dingo.core.utils import (
     set_requires_grad_flag,
@@ -32,10 +31,59 @@
 from dingo.gw.dataset import WaveformDataset
 from dingo.core.posterior_models import BasePosteriorModel
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
+def _standardization_is_missing(standardization: dict) -> bool:
+    if standardization in (None, "???"):
+        return True
+    return (
+        standardization.get("mean") in (None, "???")
+        or standardization.get("std") in (None, "???")
+    )
+
+
+def _populate_parameter_standardization(
+    wfd: WaveformDataset,
+    data_settings: dict,
+    asd_dataset_config: dict,
+) -> None:
+    if not _standardization_is_missing(data_settings.get("standardization")):
+        logger.info("Using previously-calculated parameter standardizations.")
+        _sync_parameter_standardization_transform_configs(data_settings)
+        return
+
+    logger.info("Calculating new parameter standardizations.")
+    asd_dataset = instantiate(asd_dataset_config, domain_update=wfd.domain.domain_dict)
+    runtime_dependencies = {
+        "domain": wfd.domain,
+        "asd_dataset": asd_dataset,
+    }
+    transforms = [
+        instantiate_with_runtime_dependencies(transform_config, runtime_dependencies)
+        for transform_config in data_settings["standardization_transforms"]
+    ]
+    selected_parameters = (
+        data_settings["inference_parameters"] + data_settings["context_parameters"]
+    )
+    data_settings["standardization"] = get_standardization_dict(
+        instantiate(data_settings["extrinsic_prior"]),
+        wfd,
+        selected_parameters,
+        torchvision.transforms.Compose(transforms),
+    )
+    _sync_parameter_standardization_transform_configs(data_settings)
+
+
+def _sync_parameter_standardization_transform_configs(data_settings: dict) -> None:
+    for transform_config in data_settings["transforms"]:
+        if transform_config.get("_target_", "").endswith(
+            "SelectStandardizeRepackageParameters"
+        ):
+            transform_config["standardization_dict"] = data_settings["standardization"]
+
+
 def copy_files_to_local(
     file_path: str,
     local_dir: Optional[str],
@@ -66,16 +114,16 @@ def copy_files_to_local(
     if local_dir is not None:
         file_name = file_path.split("/")[-1]
         local_file_path = os.path.join(local_dir, file_name)
-        log.info(f"Copying file to {local_file_path}")
+        logger.info(f"Copying file to {local_file_path}")
         # Copy file
         start_time = time.time()
         shutil.copy(file_path, local_file_path)
         elapsed_time = time.time() - start_time
-        log.info(
+        logger.info(
             "Done. This took {:2.0f}:{:2.0f} min.".format(*divmod(elapsed_time, 60))
         )
     elif leave_keys_on_disk and is_condor:
-        log.warning(
+        logger.warning(
             f"leave_waveforms_on_disk defaults to True, but local_cache_path is not specified. "
             f"This means that the waveforms will be loaded during training from {local_file_path}. "
             f"This can lead to unexpected long times for data loading during training due to network traffic. "
@@ -112,29 +160,36 @@ def prepare_training_new(
     """
     data_settings = deepcopy(train_settings["data"])
     # Optionally copy files to local and update path
-    data_settings["waveform_dataset_path"] = copy_files_to_local(
-        file_path=data_settings["waveform_dataset_path"],
+    data_settings["waveform_dataset"]["file_name"] = copy_files_to_local(
+        file_path=data_settings["waveform_dataset"]["file_name"],
         local_dir=local_settings.get("local_cache_path", None),
         leave_keys_on_disk=local_settings.get("leave_waveforms_on_disk", True),
         is_condor=True if "condor" in local_settings else False,
     )
-    wfd = build_dataset(
-        data_settings=data_settings,
+    wfd = instantiate(
+        data_settings["waveform_dataset"],
         leave_waveforms_on_disk=local_settings.get("leave_waveforms_on_disk", True),
     )  # No transforms yet
+    train_settings["data"] = data_settings
     initial_weights = {}
 
+    _populate_parameter_standardization(
+        wfd,
+        train_settings["data"],
+        train_settings["training"]["stage_0"]["asd_dataset"],
+    )
+
     # The embedding network is assumed to have an SVD projection layer. If other types
     # of embedding networks are added in the future, update this code.
 
     svd_settings = (train_settings["model"].get("embedding_kwargs") or {}).get("svd")
     if svd_settings and svd_settings.get("num_training_samples", 0) > 0:
         # First, build the SVD for seeding the embedding network.
-        log.info("\nBuilding SVD for initialization of embedding network.")
+        logger.info("\nBuilding SVD for initialization of embedding network.")
         initial_weights["V_rb_list"] = build_svd_for_embedding_network(
             wfd,
             train_settings["data"],
-            train_settings["training"]["stage_0"]["asd_dataset_path"],
+            train_settings["training"]["stage_0"]["asd_dataset"],
             num_workers=local_settings["num_workers"],
             batch_size=train_settings["training"]["stage_0"]["batch_size"],
             out_dir=train_dir,
@@ -151,19 +206,33 @@ def prepare_training_new(
     set_train_transforms(
         wfd,
         train_settings["data"],
-        train_settings["training"]["stage_0"]["asd_dataset_path"],
+        train_settings["training"]["stage_0"]["asd_dataset"],
     )
 
-    # This modifies the model settings in-place.
-    autocomplete_model_kwargs(train_settings["model"], wfd[0])
+    data_sample = wfd[0]
+    train_settings["model"]["embedding_kwargs"]["input_dims"] = list(
+        data_sample[1].shape
+    )
+    train_settings["model"]["posterior_kwargs"]["input_dim"] = len(data_sample[0])
+    try:
+        gnpe_proxy_dim = len(data_sample[2])
+        train_settings["model"]["embedding_kwargs"]["added_context"] = True
+        train_settings["model"]["posterior_kwargs"]["context_dim"] = (
+            train_settings["model"]["embedding_kwargs"]["output_dim"] + gnpe_proxy_dim
+        )
+    except IndexError:
+        train_settings["model"]["embedding_kwargs"]["added_context"] = False
+        train_settings["model"]["posterior_kwargs"]["context_dim"] = train_settings[
+            "model"
+        ]["embedding_kwargs"]["output_dim"]
     full_settings = {
         "dataset_settings": wfd.settings,
         "train_settings": train_settings,
     }
 
-    log.info("\nInitializing new posterior model.")
-    log.info("Complete settings:")
-    log.info(yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
+    logger.info("\nInitializing new posterior model.")
+    logger.info("Complete settings:")
+    logger.info(yaml.dump(full_settings, default_flow_style=False, sort_keys=False))
 
     pm = build_model_from_kwargs(
         settings=full_settings,
@@ -181,7 +250,7 @@ def prepare_training_new(
                 **local_settings["wandb"],
             )
         except ImportError:
-            log.warning("WandB is enabled but not installed.")
+            logger.warning("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -213,14 +282,14 @@ def prepare_training_resume(
     )
     data_settings = deepcopy(pm.metadata["train_settings"]["data"])
     # Optionally copy files to local and update path
-    data_settings["waveform_dataset_path"] = copy_files_to_local(
-        file_path=data_settings["waveform_dataset_path"],
+    data_settings["waveform_dataset"]["file_name"] = copy_files_to_local(
+        file_path=data_settings["waveform_dataset"]["file_name"],
         local_dir=local_settings.get("local_cache_path", None),
         leave_keys_on_disk=local_settings.get("leave_waveforms_on_disk", True),
         is_condor=True if "condor" in local_settings else False,
     )
-    wfd = build_dataset(
-        data_settings=data_settings,
+    wfd = instantiate(
+        data_settings["waveform_dataset"],
         leave_waveforms_on_disk=local_settings.get("leave_waveforms_on_disk", True),
     )
 
@@ -234,7 +303,7 @@ def prepare_training_resume(
                 **local_settings["wandb"],
             )
         except ImportError:
-            log.warning("WandB is enabled but not installed.")
+            logger.warning("WandB is enabled but not installed.")
 
     return pm, wfd
 
@@ -273,7 +342,7 @@ def initialize_stage(
     train_settings = pm.metadata["train_settings"]
 
     # Rebuild transforms based on possibly different noise.
-    set_train_transforms(wfd, train_settings["data"], stage["asd_dataset_path"])
+    set_train_transforms(wfd, train_settings["data"], stage["asd_dataset"])
 
     # Allows for changes in batch size between stages.
     train_loader, test_loader = build_train_and_test_loaders(
@@ -286,7 +355,7 @@ def initialize_stage(
     if not resume:
         # New optimizer and scheduler. If we are resuming, these should have been
         # loaded from the checkpoint.
-        log.info("Initializing new optimizer and scheduler.")
+        logger.info("Initializing new optimizer and scheduler.")
         pm.optimizer_kwargs = stage["optimizer"]
         pm.scheduler_kwargs = stage["scheduler"]
         pm.initialize_optimizer_and_scheduler()
@@ -303,7 +372,7 @@ def initialize_stage(
             )
     n_grad = get_number_of_model_parameters(pm.network, (True,))
     n_nograd = get_number_of_model_parameters(pm.network, (False,))
-    log.info(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}\n")
+    logger.info(f"Fixed parameters: {n_nograd}\nLearnable parameters: {n_grad}\n")
 
     return train_loader, test_loader
 
@@ -351,14 +420,14 @@ def train_stages(
         stage = stages[n]
 
         if pm.epoch == end_epochs[n] - stage["epochs"]:
-            log.info(f"\nBeginning training stage {n}. Settings:")
-            log.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            logger.info(f"\nBeginning training stage {n}. Settings:")
+            logger.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=False
             )
         else:
-            log.info(f"\nResuming training in stage {n}. Settings:")
-            log.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
+            logger.info(f"\nResuming training in stage {n}. Settings:")
+            logger.info(yaml.dump(stage, default_flow_style=False, sort_keys=False))
             train_loader, test_loader = initialize_stage(
                 pm, wfd, stage, local_settings["num_workers"], resume=True
             )
@@ -367,7 +436,7 @@ def train_stages(
             try:
                 early_stopping = EarlyStopping(**stage["early_stopping"])
             except Exception:
-                log.warning(
+                logger.warning(
                     "Early stopping settings invalid. Please pass 'patience', 'delta', 'metric'"
                 )
                 raise
@@ -389,10 +458,10 @@ def train_stages(
 
         if pm.epoch == end_epochs[n]:
             save_file = os.path.join(train_dir, f"model_stage_{n}.pt")
-            log.info(f"Training stage complete. Saving to {save_file}.")
+            logger.info(f"Training stage complete. Saving to {save_file}.")
             pm.save_model(save_file, save_training_info=True)
         if runtime_limits.local_limits_exceeded(pm.epoch):
-            log.info("Local runtime limits reached. Ending program.")
+            logger.info("Local runtime limits reached. Ending program.")
             break
 
     if pm.epoch == end_epochs[-1]:
@@ -402,12 +471,14 @@ def train_stages(
 
 
 def _resolve_training_input_paths(train_settings: dict) -> None:
-    train_settings["data"]["waveform_dataset_path"] = to_absolute_path(
-        train_settings["data"]["waveform_dataset_path"]
+    train_settings["data"]["waveform_dataset"]["file_name"] = to_absolute_path(
+        train_settings["data"]["waveform_dataset"]["file_name"]
     )
     for stage in train_settings["training"].values():
-        if isinstance(stage, dict) and "asd_dataset_path" in stage:
-            stage["asd_dataset_path"] = to_absolute_path(stage["asd_dataset_path"])
+        if isinstance(stage, dict) and "asd_dataset" in stage:
+            stage["asd_dataset"]["file_name"] = to_absolute_path(
+                stage["asd_dataset"]["file_name"]
+            )
 
 
 @hydra.main(
@@ -424,7 +495,7 @@ def train_local(cfg: DictConfig):
     os.makedirs(train_dir, exist_ok=True)
 
     if checkpoint is None:
-        log.info("Beginning new training run.")
+        logger.info("Beginning new training run.")
         train_settings = settings
         _resolve_training_input_paths(train_settings)
 
@@ -443,7 +514,7 @@ def train_local(cfg: DictConfig):
 
                     local_settings["wandb"]["id"] = wandb.util.generate_id()
                 except ImportError:
-                    log.warning("wandb not installed, cannot generate run id.")
+                    logger.warning("wandb not installed, cannot generate run id.")
             yaml.dump(local_settings, f, default_flow_style=False, sort_keys=False)
 
         pm, wfd = prepare_training_new(train_settings, train_dir, local_settings)
@@ -452,7 +523,7 @@ def train_local(cfg: DictConfig):
         checkpoint = to_absolute_path(checkpoint)
         if not os.path.isfile(checkpoint):
             raise FileNotFoundError(f"Checkpoint not found: {checkpoint}")
-        log.info("Resuming training run.")
+        logger.info("Resuming training run.")
         with open(os.path.join(train_dir, "local_settings.yaml"), "r") as f:
             local_settings = yaml.safe_load(f)
         pm, wfd = prepare_training_resume(checkpoint, local_settings, train_dir)
@@ -462,11 +533,11 @@ def train_local(cfg: DictConfig):
 
     if complete:
         if exit_command:
-            log.info(
+            logger.info(
                 f"All training stages complete. Executing exit command: {exit_command}."
             )
             os.system(exit_command)
         else:
-            log.info("All training stages complete.")
+            logger.info("All training stages complete.")
     else:
-        log.info("Program terminated due to runtime limit.")
+        logger.info("Program terminated due to runtime limit.")
diff --git a/dingo/gw/training/train_pipeline_condor.py b/dingo/gw/training/train_pipeline_condor.py
index 06550f5a2..4bde7af41 100644
--- a/dingo/gw/training/train_pipeline_condor.py
+++ b/dingo/gw/training/train_pipeline_condor.py
@@ -1,16 +1,19 @@
 import os
 import sys
+import logging
 from os.path import join, isfile
 import yaml
 import argparse
 
-from dingo.core.utils.logging_utils import logger, setup_logger
+from dingo.core.utils.logging_utils import setup_logger
 from dingo.gw.training import (
     prepare_training_new,
     prepare_training_resume,
     train_stages,
 )
 
+logger = logging.getLogger(__name__)
+
 
 def create_submission_file(
     train_dir: str, condor_settings: dict, filename: str = "submission_file.sub"
diff --git a/dingo/gw/training/utils.py b/dingo/gw/training/utils.py
index 8e9736280..d864c1852 100644
--- a/dingo/gw/training/utils.py
+++ b/dingo/gw/training/utils.py
@@ -9,7 +9,7 @@
 
 from dingo.core.utils.backward_compatibility import torch_load_with_fallback
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 logging.captureWarnings(True)
 
 
@@ -30,7 +30,7 @@ def append_stage(cfg: DictConfig):
         if k.startswith("stage_")
     ]
     num_stages = len(stages)
-    log.info(f"Checkpoint training plan consists of {num_stages} stages.")
+    logger.info(f"Checkpoint training plan consists of {num_stages} stages.")
 
     new_stage = cfg["stage"]
 
@@ -43,20 +43,20 @@ def append_stage(cfg: DictConfig):
         current_epoch = d["epoch"]
         stage_epoch = np.sum([s["epochs"] for s in stages[: cfg["replace"]]])
         if current_epoch > stage_epoch:
-            log.info(
+            logger.info(
                 f"WARNING: Modification to training plan changes a training stage "
                 f"that has already started. Current model epoch is {current_epoch}. "
                 f"Proceed at your own risk!"
             )
-        log.info(f"Replacing planned stage {cfg['replace']} with new stage.")
+        logger.info(f"Replacing planned stage {cfg['replace']} with new stage.")
         new_stage_number = cfg["replace"]
     else:
-        log.info("Appending new stage to training plan.")
+        logger.info("Appending new stage to training plan.")
         new_stage_number = num_stages
 
     d["metadata"]["train_settings"]["training"][f"stage_{new_stage_number}"] = new_stage
-    log.info("Summary of new training plan:")
-    log.info(
+    logger.info("Summary of new training plan:")
+    logger.info(
         yaml.dump(
             d["metadata"]["train_settings"]["training"],
             default_flow_style=False,
diff --git a/dingo/gw/transforms/noise_transforms.py b/dingo/gw/transforms/noise_transforms.py
index d864980eb..eb9f382ca 100644
--- a/dingo/gw/transforms/noise_transforms.py
+++ b/dingo/gw/transforms/noise_transforms.py
@@ -137,7 +137,9 @@ class WhitenAndScaleStrain(object):
     This accounts for frequency binning 
     """
 
-    def __init__(self, scale_factor):
+    def __init__(self, scale_factor=None, domain=None):
+        if domain is not None:
+            scale_factor = domain.noise_std
         self.scale_factor = scale_factor
 
     def __call__(self, input_sample):
@@ -195,8 +197,10 @@ class RepackageStrainsAndASDS(object):
         i = 2: 1 / (asd * 1e23)
     """
 
-    def __init__(self, ifos, first_index=0):
+    def __init__(self, ifos, first_index=0, domain=None):
         self.ifos = ifos
+        if domain is not None:
+            first_index = domain.min_idx
         self.first_index = first_index
 
     def __call__(self, input_sample):
diff --git a/dingo/gw/transforms/parameter_transforms.py b/dingo/gw/transforms/parameter_transforms.py
index aff5889aa..4ab142bbf 100644
--- a/dingo/gw/transforms/parameter_transforms.py
+++ b/dingo/gw/transforms/parameter_transforms.py
@@ -12,7 +12,10 @@ class SampleExtrinsicParameters(object):
 
     def __init__(self, extrinsic_prior_dict):
         self.extrinsic_prior_dict = extrinsic_prior_dict
-        self.prior = BBHExtrinsicPriorDict(extrinsic_prior_dict)
+        if isinstance(extrinsic_prior_dict, BBHExtrinsicPriorDict):
+            self.prior = extrinsic_prior_dict
+        else:
+            self.prior = BBHExtrinsicPriorDict(extrinsic_prior_dict)
 
     def __call__(self, input_sample):
         sample = input_sample.copy()
diff --git a/dingo/gw/transforms/waveform_transforms.py b/dingo/gw/transforms/waveform_transforms.py
index 7bba751ef..6b59847ad 100644
--- a/dingo/gw/transforms/waveform_transforms.py
+++ b/dingo/gw/transforms/waveform_transforms.py
@@ -5,7 +5,7 @@
 
 from dingo.gw.domains import MultibandedFrequencyDomain, UniformFrequencyDomain
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class DecimateAll(object):
@@ -440,14 +440,14 @@ def __init__(
         maximum_frequency: Optional[float | dict[str, float]]
             Update of f_max. If a float, the same value will be used for all detectors.
         print_output: bool
-            Whether to write settings information to the log.
+            Whether to write settings information to the logger.
         """
         self.sample_frequencies = domain.sample_frequencies
         self.minimum_frequency = minimum_frequency
         self.maximum_frequency = maximum_frequency
 
         if print_output:
-            log.info(
+            logger.info(
                 f"Transform MaskDataForFrequencyRangeUpdate activated:"
                 f"  Settings: \n"
                 f"    - Minimum_frequency update: {self.minimum_frequency}\n"
diff --git a/dingo/gw/waveform_generator/waveform_generator.py b/dingo/gw/waveform_generator/waveform_generator.py
index ec6c01d96..0be82ab38 100644
--- a/dingo/gw/waveform_generator/waveform_generator.py
+++ b/dingo/gw/waveform_generator/waveform_generator.py
@@ -32,7 +32,7 @@
 )
 from dingo.gw.transforms.waveform_transforms import DecimateAll
 
-log = logging.getLogger(__name__)
+logger = logging.getLogger(__name__)
 
 
 class WaveformGenerator:
@@ -145,12 +145,12 @@ def spin_conversion_phase(self):
     @spin_conversion_phase.setter
     def spin_conversion_phase(self, value):
         if value is None:
-            log.info(
+            logger.info(
                 "Setting spin_conversion_phase = None. Using phase parameter for "
                 "conversion to cartesian spins."
             )
         else:
-            log.info(
+            logger.info(
                 f"Setting spin_conversion_phase = {value}. Using this value for the "
                 f"phase parameter for conversion to cartesian spins."
             )
@@ -586,7 +586,7 @@ def generate_FD_waveform(
         # numbers if multibanding is used. If that happens, turn off multibanding to
         # fix this.
         if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-            log.warning(
+            logger.warning(
                 f"Generation with parameters {parameters_lal} likely numerically "
                 f"unstable due to multibanding, turn off multibanding."
             )
@@ -609,7 +609,7 @@ def generate_FD_waveform(
                 *parameters_lal[lal_dict_idx + 1 :],
             )
             if max(np.max(np.abs(hp.data.data)), np.max(np.abs(hc.data.data))) > 1e-20:
-                log.warning(
+                logger.warning(
                     f"Turning off multibanding for parameters {parameters_lal}"
                     f" likely numerically might not have fixed it, check manually."
                 )
@@ -1569,8 +1569,8 @@ def sum_contributions_m(x_m, phase_shift=0.0):
 if __name__ == "__main__":
     import pandas as pd
     import numpy as np
+    from bilby.gw.prior import BBHPriorDict
     from dingo.gw.domains import build_domain
-    from dingo.gw.prior import build_prior_with_defaults
 
     domain_settings = {
         "type": "UniformFrequencyDomain",
@@ -1595,7 +1595,7 @@ def sum_contributions_m(x_m, phase_shift=0.0):
         "phi_jl": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic")',
         "geocent_time": 0.0,
     }
-    prior = build_prior_with_defaults(intrinsic_dict)
+    prior = BBHPriorDict(intrinsic_dict)
     p = prior.sample()
     p = {
         "mass_ratio": 0.3501852584069329,
@@ -1624,7 +1624,7 @@ def sum_contributions_m(x_m, phase_shift=0.0):
     pol_m = wfg.generate_hplus_hcross_m(p)
 
     phase_shift = np.random.uniform(high=2 * np.pi)
-    log.info(f"{phase_shift:.2f}")
+    logger.info(f"{phase_shift:.2f}")
     pol = sum_contributions_m(pol_m, phase_shift=phase_shift)
 
     pol_ref = wfg.generate_hplus_hcross({**p, "phase": p["phase"] + phase_shift})
diff --git a/tests/core/test_posterior_model.py b/tests/core/test_posterior_model.py
index 1feb4275d..263143095 100644
--- a/tests/core/test_posterior_model.py
+++ b/tests/core/test_posterior_model.py
@@ -44,7 +44,7 @@ def data_setup_pm_1():
     }
 
     d.model_kwargs = {
-        "posterior_model_type": "normalizing_flow",
+        "_target_": "dingo.core.posterior_models.NormalizingFlowPosteriorModel",
         "posterior_kwargs": d.posterior_kwargs,
         "embedding_kwargs": d.embedding_kwargs,
     }
@@ -59,22 +59,31 @@ def data_setup_optimizer_scheduler():
     d = types.SimpleNamespace()
 
     d.adam_kwargs = {
-        "type": "adam",
+        "_target_": "torch.optim.Adam",
+        "_partial_": True,
         "lr": 0.0001,
     }
     d.sgd_kwargs = {
-        "type": "sgd",
+        "_target_": "torch.optim.SGD",
+        "_partial_": True,
         "lr": 0.0001,
         "momentum": 0.9,
     }
 
-    d.step_kwargs = {"type": "step", "step_size": 3, "gamma": 0.5}
+    d.step_kwargs = {
+        "_target_": "torch.optim.lr_scheduler.StepLR",
+        "_partial_": True,
+        "step_size": 3,
+        "gamma": 0.5,
+    }
     d.cosine_kwargs = {
-        "type": "cosine",
+        "_target_": "torch.optim.lr_scheduler.CosineAnnealingLR",
+        "_partial_": True,
         "T_max": 10,
     }
     d.rop_kwargs = {
-        "type": "reduce_on_plateau",
+        "_target_": "torch.optim.lr_scheduler.ReduceLROnPlateau",
+        "_partial_": True,
         "factor": 0.5,
         "patience": 3,
     }
@@ -115,6 +124,7 @@ def test_pm_saving_and_loading_basic(data_setup_pm_1):
 
     # load saved model
     pm_1 = NormalizingFlowPosteriorModel(model_filename=d.model_filename, device="cpu")
+    assert pm_1.metadata.train_settings.model._target_ == d.model_kwargs["_target_"]
 
     # build a model with identical kwargs
     pm_2 = NormalizingFlowPosteriorModel(metadata=d.metadata, device="cpu")
@@ -222,3 +232,22 @@ def test_pm_scheduler(data_setup_pm_1, data_setup_optimizer_scheduler):
         factors.append(lr / pm.optimizer.defaults["lr"])
         torchutils.perform_scheduler_step(pm.scheduler, loss)
     assert np.allclose(factors, e.cosine_factors), "Scheduler does not load correctly."
+
+
+def test_pm_hydra_optimizer_scheduler(data_setup_pm_1):
+    d = data_setup_pm_1
+    pm = NormalizingFlowPosteriorModel(metadata=d.metadata, device="cpu")
+    pm.optimizer_kwargs = {
+        "_target_": "torch.optim.Adam",
+        "_partial_": True,
+        "lr": 0.0001,
+    }
+    pm.scheduler_kwargs = {
+        "_target_": "torch.optim.lr_scheduler.CosineAnnealingLR",
+        "_partial_": True,
+        "T_max": 10,
+    }
+    pm.initialize_optimizer_and_scheduler()
+
+    assert isinstance(pm.optimizer, torch.optim.Adam)
+    assert isinstance(pm.scheduler, torch.optim.lr_scheduler.CosineAnnealingLR)
diff --git a/tests/gw/conversion/test_time_delay_from_geocenter.py b/tests/gw/conversion/test_time_delay_from_geocenter.py
index 73884351e..1471a4801 100644
--- a/tests/gw/conversion/test_time_delay_from_geocenter.py
+++ b/tests/gw/conversion/test_time_delay_from_geocenter.py
@@ -5,13 +5,19 @@
 from bilby.gw.detector import InterferometerList
 
 from dingo.gw.transforms import time_delay_from_geocenter
-from dingo.gw.prior import BBHExtrinsicPriorDict, default_extrinsic_dict
+from dingo.gw.prior import BBHExtrinsicPriorDict
+
+
+DEFAULT_EXTRINSIC_DICT = {
+    "dec": "bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')",
+    "ra": 'bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")',
+}
 
 
 @pytest.fixture
 def ifo_list_and_prior_and_tref():
     ifo_list = InterferometerList(["H1", "L1", "V1"])
-    prior = BBHExtrinsicPriorDict({k: default_extrinsic_dict[k] for k in ["ra", "dec"]})
+    prior = BBHExtrinsicPriorDict(DEFAULT_EXTRINSIC_DICT)
     tref = random.randint(
         946339215, 1577491218
     )  # random gps time between 2010 and 2030
diff --git a/tests/gw/test_mfd.py b/tests/gw/test_mfd.py
index be15f22ad..dab4db9b6 100644
--- a/tests/gw/test_mfd.py
+++ b/tests/gw/test_mfd.py
@@ -67,6 +67,25 @@ def test_mfd_domain_dict(mfd):
     np.testing.assert_equal(mfd.__dict__, domain2.__dict__)
 
 
+def test_mfd_hydra_target(mfd_params):
+    settings = {
+        "_target_": "dingo.gw.domains.MultibandedFrequencyDomain",
+        **mfd_params,
+        "base_domain": {
+            "_target_": "dingo.gw.domains.UniformFrequencyDomain",
+            **mfd_params["base_domain"],
+        },
+    }
+    settings["base_domain"].pop("type")
+
+    domain = build_domain(settings)
+    domain_expected = MultibandedFrequencyDomain(**mfd_params)
+
+    assert isinstance(domain, MultibandedFrequencyDomain)
+    assert type(domain.base_domain) is type(domain_expected.base_domain)
+    np.testing.assert_equal(domain.domain_dict, domain_expected.domain_dict)
+
+
 def test_mfd_set_new_range(mfd_params, mfd):
     domain = mfd
     # test that ValueErrors are raised for infeasible inputs
@@ -182,4 +201,3 @@ def test_mfd_time_translation_torch(mfd):
     # TODO: Is there a way to improve on this?
     assert torch.allclose(result[..., 1, :], torch.tensor(0.0), atol=1e-2)
     assert torch.allclose(result[..., 2, :], torch.tensor(constant_value))
-
diff --git a/tests/gw/test_prior_split.py b/tests/gw/test_prior_split.py
index 3a307ee2e..741df5da1 100644
--- a/tests/gw/test_prior_split.py
+++ b/tests/gw/test_prior_split.py
@@ -1,6 +1,15 @@
 import numpy as np
 import pytest
-from dingo.gw.prior import BBHExtrinsicPriorDict, BBHPriorDict, default_extrinsic_dict, default_intrinsic_dict
+from dingo.gw.prior import BBHExtrinsicPriorDict, BBHPriorDict
+
+
+DEFAULT_EXTRINSIC_DICT = {
+    "dec": "bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')",
+    "ra": 'bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")',
+    "geocent_time": "bilby.core.prior.Uniform(minimum=-0.1, maximum=0.1, name='geocent_time')",
+    "psi": 'bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")',
+    "luminosity_distance": "bilby.core.prior.Uniform(minimum=100.0, maximum=6000.0, name='luminosity_distance')",
+}
 
 
 def test_prior_constraint():
@@ -18,7 +27,7 @@ def test_mean_std():
     num_samples = 100000
     eps = 0.01
     keys = ['ra', 'dec', 'luminosity_distance']
-    prior = BBHExtrinsicPriorDict(default_extrinsic_dict)
+    prior = BBHExtrinsicPriorDict(DEFAULT_EXTRINSIC_DICT)
     mean_exact, std_exact = prior.mean_std(keys)
     mean_approx, std_approx = prior.mean_std(keys,
                                              sample_size=num_samples,
diff --git a/tests/gw/test_ufd.py b/tests/gw/test_ufd.py
index 5e4157f94..58c0b1d04 100644
--- a/tests/gw/test_ufd.py
+++ b/tests/gw/test_ufd.py
@@ -40,6 +40,19 @@ def test_FD_domain_dict(uniform_FD_params):
     assert domain.__dict__ == domain2.__dict__
 
 
+def test_FD_hydra_target(uniform_FD_params):
+    settings = {
+        "_target_": "dingo.gw.domains.UniformFrequencyDomain",
+        **uniform_FD_params,
+    }
+    domain = build_domain(settings)
+    assert isinstance(domain, UniformFrequencyDomain)
+    assert domain.domain_dict == {
+        "type": "UniformFrequencyDomain",
+        **uniform_FD_params,
+    }
+
+
 def test_FD_update_data(uniform_FD_params):
     p = uniform_FD_params
     domain = UniformFrequencyDomain(**p)
@@ -163,4 +176,3 @@ def test_FD_caching(uniform_FD_params):
     # after clearing the cache, the __call__ method should return the correct
     # result
     assert len(domain()) < len(domain_ref())
-
diff --git a/tests/gw/test_waveform_dataset.py b/tests/gw/test_waveform_dataset.py
index d5150f5c6..9fa129fdd 100644
--- a/tests/gw/test_waveform_dataset.py
+++ b/tests/gw/test_waveform_dataset.py
@@ -1,58 +1,55 @@
 import os
 import tempfile
 import uuid
+import copy
 from pathlib import Path
 from typing import Generator
-import yaml
 
 import numpy as np
 import pytest
 import bilby
 
-from dingo.gw.dataset.generate_dataset import _generate_dataset_main
 from dingo.gw.dataset.waveform_dataset import WaveformDataset
 from dingo.gw.domains import Domain
 from dingo.gw.dataset import generate_dataset
 
-SETTINGS_YAML_SMALL = """\
-# settings for domain of waveforms
-domain:
-  type: UniformFrequencyDomain
-  f_min: 10.0
-  f_max: 1024.0
-  delta_f: 1.0
-
-# settings for waveform generator
-waveform_generator:
-  approximant: IMRPhenomPv2
-  f_ref: 20.0
-
-# settings for intrinsic prior over parameters
-intrinsic_prior:
-  # prior for non-fixed parameters
-  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
-  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
-  mass_ratio: bilby.core.prior.Uniform(minimum=0.125, maximum=1.0, name='mass_ratio')
-  chirp_mass: bilby.core.prior.Uniform(minimum=25.0, maximum=100.0, name='chirp_mass')
-  phase: default
-  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
-  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
-  tilt_1: default
-  tilt_2: default
-  phi_12: default
-  phi_jl: default
-  theta_jn: default
-  # reference values for fixed (extrinsic) parameters
-  luminosity_distance: 100.0 # Mpc
-  geocent_time: 0.0 # s
-
-num_samples: 50
-
-compression:
-  svd:
-    num_training_samples: 10
-    size: 5
-"""
+SETTINGS_SMALL = {
+    "domain": {
+        "_target_": "dingo.gw.domains.UniformFrequencyDomain",
+        "f_min": 10.0,
+        "f_max": 1024.0,
+        "delta_f": 1.0,
+    },
+    "waveform_generator": {
+        "_target_": "dingo.gw.waveform_generator.WaveformGenerator",
+        "approximant": "IMRPhenomPv2",
+        "f_ref": 20.0,
+        "f_start": None,
+        "spin_conversion_phase": 0.0,
+    },
+    "intrinsic_prior": {
+        "_target_": "bilby.gw.prior.BBHPriorDict",
+        "_convert_": "all",
+        "dictionary": {
+            "mass_1": "bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')",
+            "mass_2": "bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')",
+            "mass_ratio": "bilby.core.prior.Uniform(minimum=0.125, maximum=1.0, name='mass_ratio')",
+            "chirp_mass": "bilby.core.prior.Uniform(minimum=25.0, maximum=100.0, name='chirp_mass')",
+            "phase": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")',
+            "a_1": "bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')",
+            "a_2": "bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')",
+            "tilt_1": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')",
+            "tilt_2": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')",
+            "phi_12": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")',
+            "phi_jl": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")',
+            "theta_jn": "bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')",
+            "luminosity_distance": 100.0,
+            "geocent_time": 0.0,
+        },
+    },
+    "num_samples": 50,
+    "compression": None,
+}
 
 
 @pytest.fixture(scope="session")
@@ -75,17 +72,14 @@ def generate_waveform_dataset_small(temp_dir: Path) -> Path:
     in a temporary directory.
     """
 
-    # Create temp directory and settings file
+    # Create temp directory and dataset file
     path = temp_dir / "tmp_test" / str(uuid.uuid4())
     path.mkdir(parents=True)
-    settings_path = path / "settings.yaml"
     out_file = path / "waveform_dataset.hdf5"
     num_processes = 4
 
-    with open(settings_path, "w") as fp:
-        fp.writelines(SETTINGS_YAML_SMALL)
-
-    _generate_dataset_main(str(settings_path), str(out_file), num_processes)
+    dataset = generate_dataset(SETTINGS_SMALL, num_processes)
+    dataset.to_file(str(out_file))
 
     return path
 
@@ -228,8 +222,7 @@ def test_load_waveform_dataset_with_leave_polarizations_on_disk(
 
 @pytest.fixture
 def wfd_settings():
-    settings = yaml.safe_load(SETTINGS_YAML_SMALL)
-    return settings
+    return copy.deepcopy(SETTINGS_SMALL)
 
 
 class BinaryPrior(bilby.prior.Prior):
@@ -254,9 +247,9 @@ def test_wfd_size(wfd_settings: str, binary_prior: BinaryPrior):
     # changing the waveform generator settings to create a prior which will create
     # failing waveforms for a fraction of the prior. I.e. can't generate negative
     # chirp masses so the waveform generator will fail
-    wfd_settings["intrinsic_prior"]["chirp_mass"] = binary_prior
-    del wfd_settings["intrinsic_prior"]["mass_1"]
-    del wfd_settings["intrinsic_prior"]["mass_2"]
+    wfd_settings["intrinsic_prior"]["dictionary"]["chirp_mass"] = binary_prior
+    del wfd_settings["intrinsic_prior"]["dictionary"]["mass_1"]
+    del wfd_settings["intrinsic_prior"]["dictionary"]["mass_2"]
     del wfd_settings["compression"]
     wfd = generate_dataset(wfd_settings, 1)
     assert len(wfd) == wfd_settings["num_samples"]
diff --git a/tests/gw/transforms/test_batch_transforms.py b/tests/gw/transforms/test_batch_transforms.py
index e0556871b..ad782da28 100644
--- a/tests/gw/transforms/test_batch_transforms.py
+++ b/tests/gw/transforms/test_batch_transforms.py
@@ -17,11 +17,19 @@
     RepackageStrainsAndASDS,
     UnpackDict,
 )
-from dingo.gw.prior import default_extrinsic_dict
 from dingo.gw.domains import UniformFrequencyDomain
 from dingo.gw.noise.asd_dataset import ASDDataset
 
 
+DEFAULT_EXTRINSIC_DICT = {
+    "dec": "bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')",
+    "ra": 'bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")',
+    "geocent_time": "bilby.core.prior.Uniform(minimum=-0.1, maximum=0.1, name='geocent_time')",
+    "psi": 'bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")',
+    "luminosity_distance": "bilby.core.prior.Uniform(minimum=100.0, maximum=6000.0, name='luminosity_distance')",
+}
+
+
 @pytest.fixture
 def standardization_dict():
     return {
@@ -96,7 +104,7 @@ def transform_list(standardization_dict, asd_dataset, domain):
     ifo_list = InterferometerList(["H1", "L1"])
     ref_time = 1126259462.391
     transforms = [
-        SampleExtrinsicParameters(default_extrinsic_dict),
+        SampleExtrinsicParameters(DEFAULT_EXTRINSIC_DICT),
         GetDetectorTimes(ifo_list, ref_time),
         GNPECoalescenceTimes(ifo_list, Uniform(minimum=-0.001, maximum=0.001), True),
         ProjectOntoDetectors(ifo_list, domain, ref_time),
@@ -129,4 +137,4 @@ def test_unbatched_training_transforms(transform_list, input_sample_unbatched):
     transforms = torchvision.transforms.Compose(transform_list)
     output = transforms(input_sample_unbatched)
     assert output[0].ndim == 1
-    assert output[1].ndim == 3
\ No newline at end of file
+    assert output[1].ndim == 3
diff --git a/tests/gw/transforms/test_detector_projection.py b/tests/gw/transforms/test_detector_projection.py
index 267dc014f..f9bf6a82e 100644
--- a/tests/gw/transforms/test_detector_projection.py
+++ b/tests/gw/transforms/test_detector_projection.py
@@ -9,10 +9,18 @@
     SampleExtrinsicParameters,
     time_delay_from_geocenter,
 )
-from dingo.gw.prior import default_extrinsic_dict
 from dingo.gw.domains import build_domain
 
 
+DEFAULT_EXTRINSIC_DICT = {
+    "dec": "bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')",
+    "ra": 'bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")',
+    "geocent_time": "bilby.core.prior.Uniform(minimum=-0.1, maximum=0.1, name='geocent_time')",
+    "psi": 'bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")',
+    "luminosity_distance": "bilby.core.prior.Uniform(minimum=100.0, maximum=6000.0, name='luminosity_distance')",
+}
+
+
 @pytest.fixture
 def reference_data_research_code():
     dir = os.path.dirname(os.path.realpath(__file__))
@@ -30,7 +38,7 @@ def reference_data_research_code():
 @pytest.fixture
 def setup_detector_projection():
     # setup arguments
-    extrinsic_prior_dict = default_extrinsic_dict
+    extrinsic_prior_dict = DEFAULT_EXTRINSIC_DICT
     ref_time = 1126259462.391
     domain_dict = {
         "type": "UniformFrequencyDomain",
diff --git a/tests/gw/waveform_generator/test_wfg_m.py b/tests/gw/waveform_generator/test_wfg_m.py
index cc8769939..d4e32e11d 100644
--- a/tests/gw/waveform_generator/test_wfg_m.py
+++ b/tests/gw/waveform_generator/test_wfg_m.py
@@ -14,6 +14,7 @@
 import pytest
 import numpy as np
 from matplotlib import pyplot as plt
+from bilby.gw.prior import BBHPriorDict
 
 from dingo.gw.waveform_generator import (
     WaveformGenerator,
@@ -22,7 +23,6 @@
 )
 from dingo.gw.gwutils import get_mismatch
 from dingo.gw.domains import build_domain
-from dingo.gw.prior import build_prior_with_defaults
 
 
 @pytest.fixture
@@ -75,7 +75,7 @@ def intrinsic_prior(approximant):
             "chi_2": 'bilby.gw.prior.AlignedSpin(name="chi_2", a_prior=Uniform(minimum=0, maximum=0.99))',
             "geocent_time": 0.0,
         }
-    prior = build_prior_with_defaults(intrinsic_dict)
+    prior = BBHPriorDict(intrinsic_dict)
     return prior
 
 
diff --git a/tests/gw/waveform_generator/test_wfg_mfd.py b/tests/gw/waveform_generator/test_wfg_mfd.py
index 297df73bd..1157677d1 100644
--- a/tests/gw/waveform_generator/test_wfg_mfd.py
+++ b/tests/gw/waveform_generator/test_wfg_mfd.py
@@ -1,10 +1,10 @@
 import numpy as np
 import pytest
+from bilby.gw.prior import BBHPriorDict
 from scipy.interpolate import interp1d
 
 from dingo.gw.domains import MultibandedFrequencyDomain
 from dingo.gw.gwutils import get_mismatch
-from dingo.gw.prior import build_prior_with_defaults
 from dingo.gw.waveform_generator import WaveformGenerator, NewInterfaceWaveformGenerator
 
 
@@ -47,7 +47,7 @@ def intrinsic_prior(approximant):
         "phi_jl": 'bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic")',
         "geocent_time": "bilby.core.prior.Uniform(minimum=-0.1, maximum=0.1)",
     }
-    prior = build_prior_with_defaults(intrinsic_dict)
+    prior = BBHPriorDict(intrinsic_dict)
     return prior
 
 
diff --git a/tests/test_hydra_config_targets.py b/tests/test_hydra_config_targets.py
new file mode 100644
index 000000000..b0904812b
--- /dev/null
+++ b/tests/test_hydra_config_targets.py
@@ -0,0 +1,260 @@
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+import pytest
+import torch
+from bilby.gw.prior import BBHPriorDict
+from hydra import compose, initialize_config_dir
+from hydra.utils import instantiate
+from omegaconf import OmegaConf
+
+from dingo.core.utils import get_optimizer_from_kwargs, get_scheduler_from_kwargs
+from dingo.gw.SVD import SVDBasis
+from dingo.gw.dataset import WaveformDataset
+from dingo.gw.domains import (
+    MultibandedFrequencyDomain,
+    UniformFrequencyDomain,
+    build_domain,
+)
+from dingo.gw.training.train_builders import set_train_transforms
+from dingo.gw.training.train_pipeline import _populate_parameter_standardization
+from dingo.gw.waveform_generator import WaveformGenerator
+
+
+def compose_dingo_config(config_name: str, overrides: list[str] | None = None):
+    config_dir = str((Path(__file__).parents[1] / "configs").resolve())
+    with initialize_config_dir(version_base=None, config_dir=config_dir):
+        return compose(config_name=config_name, overrides=overrides or [])
+
+
+@pytest.mark.parametrize(
+    ("domain_group", "expected_type"),
+    [
+        ("uniform_frequency", UniformFrequencyDomain),
+        ("multibanded_frequency", MultibandedFrequencyDomain),
+    ],
+)
+def test_domain_config_targets_instantiate(domain_group, expected_type):
+    cfg = compose_dingo_config("generate_dataset", overrides=[f"domain={domain_group}"])
+    domain_settings = OmegaConf.to_container(cfg.domain, resolve=True)
+
+    assert isinstance(build_domain(domain_settings), expected_type)
+
+
+def test_waveform_generator_config_target_instantiates_with_domain():
+    cfg = compose_dingo_config("generate_dataset")
+    domain = instantiate(cfg.domain)
+
+    waveform_generator = instantiate(cfg.waveform_generator, domain=domain)
+
+    assert isinstance(waveform_generator, WaveformGenerator)
+    assert waveform_generator.domain is domain
+
+
+@pytest.mark.parametrize(
+    "prior_group",
+    ["default", "precessing", "precessing_multibanded_test", "fmpe"],
+)
+def test_intrinsic_prior_config_targets_instantiate(prior_group):
+    cfg = compose_dingo_config(
+        "generate_dataset", overrides=[f"intrinsic_prior={prior_group}"]
+    )
+
+    prior = instantiate(cfg.intrinsic_prior)
+
+    assert isinstance(prior, BBHPriorDict)
+    assert "chirp_mass" in prior
+    assert "mass_ratio" in prior
+
+
+@pytest.mark.parametrize("prior_group", ["default", "fmpe"])
+def test_extrinsic_prior_configs_build_prior_dict(prior_group):
+    cfg = compose_dingo_config("train", overrides=[f"extrinsic_prior={prior_group}"])
+
+    prior = instantiate(cfg.data.extrinsic_prior)
+
+    assert set(prior) == {
+        "dec",
+        "ra",
+        "geocent_time",
+        "psi",
+        "luminosity_distance",
+    }
+
+
+@pytest.mark.parametrize(
+    "model_group",
+    ["toy_npe", "npe", "gnpe", "gnpe_init", "fmpe", "unconditional_npe"],
+)
+def test_model_configs_define_posterior_model_target(model_group):
+    cfg = compose_dingo_config("train", overrides=[f"model={model_group}"])
+
+    assert cfg.model._target_.startswith("dingo.core.posterior_models.")
+    assert "posterior_model_type" not in cfg.model
+
+
+@pytest.mark.parametrize(
+    "experiment",
+    ["train_toy", "train_npe", "train_gnpe", "train_gnpe_init", "train_fmpe"],
+)
+def test_training_experiment_stage_optimizers_are_hydra_targets(experiment):
+    cfg = compose_dingo_config("train", overrides=[f"+experiment={experiment}"])
+
+    for stage_name, stage in cfg.training.items():
+        if not stage_name.startswith("stage_"):
+            continue
+        assert "_target_" in stage.optimizer
+        assert "_partial_" in stage.optimizer
+        assert "_target_" in stage.scheduler
+        assert "_partial_" in stage.scheduler
+
+
+@pytest.mark.parametrize(
+    ("experiment", "expected_targets"),
+    [
+        (
+            None,
+            [
+                "dingo.gw.transforms.SampleExtrinsicParameters",
+                "dingo.gw.transforms.GetDetectorTimes",
+            ],
+        ),
+        (
+            "train_gnpe",
+            [
+                "dingo.gw.transforms.SampleExtrinsicParameters",
+                "dingo.gw.transforms.GetDetectorTimes",
+                "dingo.gw.transforms.GNPECoalescenceTimes",
+            ],
+        ),
+    ],
+)
+def test_training_standardization_transforms_are_configured(experiment, expected_targets):
+    overrides = [] if experiment is None else [f"+experiment={experiment}"]
+    cfg = compose_dingo_config("train", overrides=overrides)
+
+    assert [t._target_ for t in cfg.data.standardization_transforms] == expected_targets
+
+    if experiment == "train_gnpe":
+        gnpe_transform = instantiate(cfg.data.standardization_transforms[-1])
+        assert list(cfg.data.context_parameters) == gnpe_transform.context_parameters
+
+
+def test_optimizer_scheduler_config_targets_instantiate():
+    cfg = compose_dingo_config("train")
+    network = torch.nn.Linear(2, 1)
+    optimizer_settings = OmegaConf.to_container(
+        cfg.training.stage_0.optimizer, resolve=True
+    )
+    scheduler_settings = OmegaConf.to_container(
+        cfg.training.stage_0.scheduler, resolve=True
+    )
+
+    optimizer = get_optimizer_from_kwargs(
+        network.parameters(),
+        **optimizer_settings,
+    )
+    scheduler = get_scheduler_from_kwargs(optimizer, **scheduler_settings)
+
+    assert isinstance(optimizer, torch.optim.Adam)
+    assert isinstance(scheduler, torch.optim.lr_scheduler.CosineAnnealingLR)
+
+
+def test_svd_basis_load_target_instantiates(tmp_path):
+    file_name = tmp_path / "svd.hdf5"
+    basis = SVDBasis()
+    basis.generate_basis(np.eye(4), 2)
+    basis.to_file(str(file_name))
+
+    loaded = instantiate(
+        {
+            "_target_": "dingo.gw.dataset.compression.load_svd_basis",
+            "file_name": str(file_name),
+        }
+    )
+
+    assert isinstance(loaded, SVDBasis)
+    assert loaded.V.shape == (4, 2)
+
+
+def test_training_standardization_is_computed_before_transform_instantiation():
+    cfg = compose_dingo_config(
+        "train",
+        overrides=[
+            "data.inference_parameters=[chirp_mass,mass_ratio]",
+            "model.embedding_kwargs.svd.num_training_samples=0",
+        ],
+    )
+    settings = OmegaConf.to_container(cfg, resolve=True)
+
+    domain_config = {
+        "_target_": "dingo.gw.domains.UniformFrequencyDomain",
+        "f_min": 0.0,
+        "f_max": 4.0,
+        "delta_f": 1.0,
+    }
+    domain = instantiate(domain_config)
+    wfd = WaveformDataset(
+        dictionary={
+            "settings": {"domain": domain_config, "compression": None},
+            "parameters": pd.DataFrame(
+                {
+                    "chirp_mass": [30.0, 31.0, 32.0],
+                    "mass_ratio": [0.5, 0.6, 0.7],
+                }
+            ),
+            "polarizations": {
+                "h_plus": np.zeros((3, len(domain)), dtype=np.complex64),
+                "h_cross": np.zeros((3, len(domain)), dtype=np.complex64),
+            },
+        }
+    )
+    settings["training"]["stage_0"]["asd_dataset"] = {
+        "_target_": "dingo.gw.noise.asd_dataset.ASDDataset",
+        "dictionary": {
+            "settings": {"domain_dict": domain.domain_dict},
+            "asds": {
+                "H1": np.ones((2, len(domain))),
+                "L1": np.ones((2, len(domain))),
+            },
+            "gps_times": {
+                "H1": np.arange(2),
+                "L1": np.arange(2),
+            },
+        },
+        "ifos": ["H1", "L1"],
+        "precision": "single",
+        "domain_update": None,
+    }
+
+    _populate_parameter_standardization(
+        wfd,
+        settings["data"],
+        settings["training"]["stage_0"]["asd_dataset"],
+    )
+    set_train_transforms(
+        wfd,
+        settings["data"],
+        settings["training"]["stage_0"]["asd_dataset"],
+    )
+
+    assert settings["data"]["standardization"]["mean"] == {
+        "chirp_mass": 31.0,
+        "mass_ratio": 0.6,
+    }
+    assert list(settings["data"]["standardization"]["std"]) == [
+        "chirp_mass",
+        "mass_ratio",
+    ]
+    assert [type(t).__name__ for t in wfd.transform.transforms] == [
+        "SampleExtrinsicParameters",
+        "GetDetectorTimes",
+        "ProjectOntoDetectors",
+        "SampleNoiseASD",
+        "WhitenAndScaleStrain",
+        "AddWhiteNoiseComplex",
+        "SelectStandardizeRepackageParameters",
+        "RepackageStrainsAndASDS",
+        "UnpackDict",
+    ]

From 2e821ae476ad6eb66d6aae7708746e7c1e65d4f1 Mon Sep 17 00:00:00 2001
From: Heinrich Campe <49278231+hcampe@users.noreply.github.com>
Date: Fri, 10 Jul 2026 16:02:41 +0200
Subject: [PATCH 15/15] simplified configs

---
 configs/append_training_stage.yaml            |  33 -----
 configs/estimate_psds.yaml                    |  33 -----
 configs/evaluate_multibanded_domain.yaml      |  13 --
 configs/experiment/asd_fiducial.yaml          |  13 --
 configs/experiment/asd_full.yaml              |  13 --
 configs/experiment/generate_fmpe_dataset.yaml |  30 ----
 configs/experiment/generate_gnpe_dataset.yaml |  30 ----
 configs/experiment/generate_npe_dataset.yaml  |  30 ----
 configs/experiment/generate_toy_dataset.yaml  |   1 -
 configs/experiment/importance_sampling.yaml   |  16 --
 configs/experiment/synthetic_asd.yaml         |  17 ---
 configs/experiment/train_fmpe.yaml            |  78 ----------
 configs/experiment/train_gnpe.yaml            | 139 ------------------
 configs/experiment/train_gnpe_init.yaml       |  70 ---------
 configs/experiment/train_npe.yaml             |  78 ----------
 configs/extrinsic_prior/default.yaml          |   8 -
 configs/extrinsic_prior/fmpe.yaml             |   8 -
 configs/generate_asd_dataset.yaml             |  12 +-
 configs/generate_dataset.yaml                 |  25 +++-
 configs/generate_synthetic_asd_dataset.yaml   |  33 -----
 configs/hydra/default.yaml                    |  11 --
 configs/importance_weights.yaml               |  47 ------
 configs/intrinsic_prior/default.yaml          |  13 --
 configs/intrinsic_prior/fmpe.yaml             |  17 ---
 configs/intrinsic_prior/precessing.yaml       |  17 ---
 .../precessing_multibanded_test.yaml          |  17 ---
 configs/merge_asd_datasets.yaml               |  35 -----
 configs/model/unconditional_npe.yaml          |  12 --
 configs/optimizer/adam.yaml                   |   3 -
 configs/scheduler/cosine.yaml                 |   3 -
 configs/train.yaml                            |  32 +++-
 configs/unconditional_density_estimation.yaml |  29 ----
 32 files changed, 59 insertions(+), 857 deletions(-)
 delete mode 100644 configs/append_training_stage.yaml
 delete mode 100644 configs/estimate_psds.yaml
 delete mode 100644 configs/evaluate_multibanded_domain.yaml
 delete mode 100644 configs/experiment/asd_fiducial.yaml
 delete mode 100644 configs/experiment/asd_full.yaml
 delete mode 100644 configs/experiment/generate_fmpe_dataset.yaml
 delete mode 100644 configs/experiment/generate_gnpe_dataset.yaml
 delete mode 100644 configs/experiment/generate_npe_dataset.yaml
 delete mode 100644 configs/experiment/importance_sampling.yaml
 delete mode 100644 configs/experiment/synthetic_asd.yaml
 delete mode 100644 configs/experiment/train_fmpe.yaml
 delete mode 100644 configs/experiment/train_gnpe.yaml
 delete mode 100644 configs/experiment/train_gnpe_init.yaml
 delete mode 100644 configs/experiment/train_npe.yaml
 delete mode 100644 configs/extrinsic_prior/default.yaml
 delete mode 100644 configs/extrinsic_prior/fmpe.yaml
 delete mode 100644 configs/generate_synthetic_asd_dataset.yaml
 delete mode 100644 configs/hydra/default.yaml
 delete mode 100644 configs/importance_weights.yaml
 delete mode 100644 configs/intrinsic_prior/default.yaml
 delete mode 100644 configs/intrinsic_prior/fmpe.yaml
 delete mode 100644 configs/intrinsic_prior/precessing.yaml
 delete mode 100644 configs/intrinsic_prior/precessing_multibanded_test.yaml
 delete mode 100644 configs/merge_asd_datasets.yaml
 delete mode 100644 configs/model/unconditional_npe.yaml
 delete mode 100644 configs/optimizer/adam.yaml
 delete mode 100644 configs/scheduler/cosine.yaml
 delete mode 100644 configs/unconditional_density_estimation.yaml

diff --git a/configs/append_training_stage.yaml b/configs/append_training_stage.yaml
deleted file mode 100644
index 1dca1ca82..000000000
--- a/configs/append_training_stage.yaml
+++ /dev/null
@@ -1,33 +0,0 @@
-defaults:
-  - hydra: default
-  - optimizer: adam
-  - scheduler: cosine
-  - _self_
-
-checkpoint: ???
-out_file: updated_model.pt
-replace: null
-
-stage:
-  optimizer: ${optimizer}
-  scheduler: ${scheduler}
-  epochs: 20
-  asd_dataset:
-    _target_: dingo.gw.noise.asd_dataset.ASDDataset
-    file_name: training_data/asd_dataset/asds_O1.hdf5
-    ifos: null
-    precision: single
-    domain_update: null
-  freeze_rb_layer: false
-  batch_size: 64
-  early_stopping: null
-
-hydra:
-  job:
-    name: append_training_stage
-    config:
-      override_dirname:
-        exclude_keys:
-          - checkpoint
-          - out_file
-          - stage.asd_dataset.file_name
diff --git a/configs/estimate_psds.yaml b/configs/estimate_psds.yaml
deleted file mode 100644
index 36e58ca2f..000000000
--- a/configs/estimate_psds.yaml
+++ /dev/null
@@ -1,33 +0,0 @@
-defaults:
-  - hydra: default
-  - _self_
-
-data_dir: asd_dataset
-time_segments_file: ???
-verbose: false
-
-dataset_settings:
-  f_min: 0
-  f_max: 2048
-  f_s: 4096
-  time_psd: 1024
-  T: 8.0
-  window:
-    roll_off: 0.4
-    type: tukey
-  time_gap: 0
-  num_psds_max: 1
-  channels: null
-  detectors:
-    - H1
-    - L1
-  observing_run: O1
-
-hydra:
-  job:
-    name: estimate_psds
-    config:
-      override_dirname:
-        exclude_keys:
-          - data_dir
-          - time_segments_file
diff --git a/configs/evaluate_multibanded_domain.yaml b/configs/evaluate_multibanded_domain.yaml
deleted file mode 100644
index f372aa01f..000000000
--- a/configs/evaluate_multibanded_domain.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
-defaults:
-  - hydra: default
-  - domain: multibanded_frequency
-  - waveform_generator: phenom_xphm
-  - intrinsic_prior: precessing_multibanded_test
-  - _self_
-
-num_samples: 1
-compression: null
-
-hydra:
-  job:
-    name: evaluate_multibanded_domain
diff --git a/configs/experiment/asd_fiducial.yaml b/configs/experiment/asd_fiducial.yaml
deleted file mode 100644
index d69d57528..000000000
--- a/configs/experiment/asd_fiducial.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
-# @package _global_
-
-dataset_settings:
-  f_s: 4096
-  time_psd: 1024
-  T: 8.0
-  window:
-    roll_off: 0.4
-    type: tukey
-  time_gap: 0
-  num_psds_max: 1
-  detectors: [H1, L1]
-  observing_run: O1
diff --git a/configs/experiment/asd_full.yaml b/configs/experiment/asd_full.yaml
deleted file mode 100644
index ca2a2704b..000000000
--- a/configs/experiment/asd_full.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
-# @package _global_
-
-dataset_settings:
-  f_s: 4096
-  time_psd: 1024
-  T: 8.0
-  window:
-    roll_off: 0.4
-    type: tukey
-  time_gap: 0
-  num_psds_max: 0
-  detectors: [H1, L1]
-  observing_run: O1
diff --git a/configs/experiment/generate_fmpe_dataset.yaml b/configs/experiment/generate_fmpe_dataset.yaml
deleted file mode 100644
index 308d51fd0..000000000
--- a/configs/experiment/generate_fmpe_dataset.yaml
+++ /dev/null
@@ -1,30 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /domain: uniform_frequency
-  - override /waveform_generator: phenom_pv2
-  - override /intrinsic_prior: fmpe
-  - _self_
-
-num_samples: 5000000
-compression:
-  - _target_: dingo.gw.transforms.WhitenFixedASD
-    _partial_: true
-    _runtime_dependencies_: [domain]
-    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
-    inverse: false
-  - _target_: dingo.gw.SVD.ApplySVD
-    svd_basis:
-      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
-      _partial_: true
-      _runtime_dependencies_:
-        - waveform_generator
-        - prior
-        - num_processes
-        - settings
-        - compression_transforms
-        - compression_settings
-      size: 200
-      num_training_samples: 50000
-      num_validation_samples: 10000
-out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_gnpe_dataset.yaml b/configs/experiment/generate_gnpe_dataset.yaml
deleted file mode 100644
index 4f54a2bb8..000000000
--- a/configs/experiment/generate_gnpe_dataset.yaml
+++ /dev/null
@@ -1,30 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /domain: uniform_frequency
-  - override /waveform_generator: phenom_xphm
-  - override /intrinsic_prior: precessing
-  - _self_
-
-num_samples: 5000000
-compression:
-  - _target_: dingo.gw.transforms.WhitenFixedASD
-    _partial_: true
-    _runtime_dependencies_: [domain]
-    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
-    inverse: false
-  - _target_: dingo.gw.SVD.ApplySVD
-    svd_basis:
-      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
-      _partial_: true
-      _runtime_dependencies_:
-        - waveform_generator
-        - prior
-        - num_processes
-        - settings
-        - compression_transforms
-        - compression_settings
-      size: 200
-      num_training_samples: 50000
-      num_validation_samples: 10000
-out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_npe_dataset.yaml b/configs/experiment/generate_npe_dataset.yaml
deleted file mode 100644
index 44d952bf8..000000000
--- a/configs/experiment/generate_npe_dataset.yaml
+++ /dev/null
@@ -1,30 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /domain: uniform_frequency
-  - override /waveform_generator: phenom_xphm
-  - override /intrinsic_prior: default
-  - _self_
-
-num_samples: 5000000
-compression:
-  - _target_: dingo.gw.transforms.WhitenFixedASD
-    _partial_: true
-    _runtime_dependencies_: [domain]
-    asd_file: aLIGO_ZERO_DET_high_P_asd.txt
-    inverse: false
-  - _target_: dingo.gw.SVD.ApplySVD
-    svd_basis:
-      _target_: dingo.gw.dataset.compression.train_svd_basis_from_waveforms
-      _partial_: true
-      _runtime_dependencies_:
-        - waveform_generator
-        - prior
-        - num_processes
-        - settings
-        - compression_transforms
-        - compression_settings
-      size: 200
-      num_training_samples: 50000
-      num_validation_samples: 10000
-out_file: training_data/waveform_dataset.hdf5
diff --git a/configs/experiment/generate_toy_dataset.yaml b/configs/experiment/generate_toy_dataset.yaml
index 386dac175..65b0281bf 100644
--- a/configs/experiment/generate_toy_dataset.yaml
+++ b/configs/experiment/generate_toy_dataset.yaml
@@ -3,7 +3,6 @@
 defaults:
   - override /domain: uniform_frequency
   - override /waveform_generator: phenom_d
-  - override /intrinsic_prior: default
   - _self_
 
 domain:
diff --git a/configs/experiment/importance_sampling.yaml b/configs/experiment/importance_sampling.yaml
deleted file mode 100644
index 0ab8eb7c2..000000000
--- a/configs/experiment/importance_sampling.yaml
+++ /dev/null
@@ -1,16 +0,0 @@
-# @package _global_
-
-num_samples: 1000
-time_marginalization:
-  n_fft: 5
-slice_plots:
-  num_slice_plots: 10
-  params_slice2d:
-    - [phase, geocent_time]
-    - [phase, tilt_1]
-calibration_marginalization:
-  num_calibration_curves: 100
-  num_calibration_nodes: 10
-  calibration_envelope:
-    H1: /path/to/H1-calibration-envelope
-    L1: /path/to/L1-calibration-envelope
diff --git a/configs/experiment/synthetic_asd.yaml b/configs/experiment/synthetic_asd.yaml
deleted file mode 100644
index 91782b5bc..000000000
--- a/configs/experiment/synthetic_asd.yaml
+++ /dev/null
@@ -1,17 +0,0 @@
-# @package _global_
-
-parameterization_settings:
-  num_spline_positions: 30
-  num_spectral_segments: 400
-  sigma: 0.14
-  delta_f: -1
-  smoothen: true
-sampling_settings:
-  bandwidth_spectral: 0.5
-  bandwidth_spline: 0.25
-  split_frequencies:
-    - 30
-    - 100
-  rescaling_asd_paths:
-    H1: /path/to/rescaling_asd_H1.hdf5
-    L1: /path/to/rescaling_asd_L1.hdf5
diff --git a/configs/experiment/train_fmpe.yaml b/configs/experiment/train_fmpe.yaml
deleted file mode 100644
index 8685b7f99..000000000
--- a/configs/experiment/train_fmpe.yaml
+++ /dev/null
@@ -1,78 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /extrinsic_prior: fmpe
-  - override /model: fmpe
-  - _self_
-
-data:
-  waveform_dataset:
-    file_name: training_data/waveform_dataset.hdf5
-    domain_update:
-      f_min: 20.0
-      f_max: 512.0
-    svd_size_update: 150
-  train_fraction: 0.95
-  detectors: [H1, L1]
-  ref_time: 1126259462.391
-  inference_parameters:
-    - chirp_mass
-    - mass_ratio
-    - a_1
-    - a_2
-    - tilt_1
-    - tilt_2
-    - phi_12
-    - phi_jl
-    - theta_jn
-    - luminosity_distance
-    - geocent_time
-    - ra
-    - dec
-    - psi
-
-training:
-  evaluate: true
-  stage_0:
-    epochs: 400
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: true
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      lr: 0.0005
-    scheduler:
-      T_max: 400
-  stage_1:
-    epochs: 100
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset/asds_O1.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: true
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      _target_: torch.optim.Adam
-      _partial_: true
-      lr: 1.0e-5
-    scheduler:
-      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
-      _partial_: true
-      T_max: 100
-
-local:
-  device: cuda
-  num_workers: 31
-  runtime_limits:
-    max_time_per_run: 360000
-    max_epochs_per_run: 500
-  checkpoint_epochs: 200
-  leave_waveforms_on_disk: true
diff --git a/configs/experiment/train_gnpe.yaml b/configs/experiment/train_gnpe.yaml
deleted file mode 100644
index a934225e0..000000000
--- a/configs/experiment/train_gnpe.yaml
+++ /dev/null
@@ -1,139 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /model: gnpe
-  - _self_
-
-data:
-  waveform_dataset:
-    file_name: training_data/waveform_dataset.hdf5
-    domain_update:
-      f_min: 20.0
-      f_max: 1024.0
-    svd_size_update: 200
-  train_fraction: 0.95
-  detectors: [H1, L1]
-  ref_time: 1126259462.391
-  gnpe_time_shifts:
-    _target_: dingo.gw.transforms.GNPECoalescenceTimes
-    ifo_list:
-      _target_: bilby.gw.detector.InterferometerList
-      interferometers: ${data.detectors}
-    kernel: bilby.core.prior.Uniform(minimum=-0.001, maximum=0.001)
-    exact_global_equivariance: true
-    inference: false
-  context_parameters:
-    - L1_time_proxy_relative
-  selected_keys:
-    - inference_parameters
-    - waveform
-    - context_parameters
-  inference_parameters:
-    - chirp_mass
-    - mass_ratio
-    - a_1
-    - a_2
-    - tilt_1
-    - tilt_2
-    - phi_12
-    - phi_jl
-    - theta_jn
-    - luminosity_distance
-    - geocent_time
-    - ra
-    - dec
-    - psi
-  standardization_transforms:
-    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
-      extrinsic_prior_dict: ${data.extrinsic_prior}
-    - _target_: dingo.gw.transforms.GetDetectorTimes
-      ifo_list:
-        _target_: bilby.gw.detector.InterferometerList
-        interferometers: ${data.detectors}
-      ref_time: ${data.ref_time}
-    - ${data.gnpe_time_shifts}
-  transforms:
-    - _target_: dingo.gw.transforms.SampleExtrinsicParameters
-      extrinsic_prior_dict: ${data.extrinsic_prior}
-    - _target_: dingo.gw.transforms.GetDetectorTimes
-      ifo_list:
-        _target_: bilby.gw.detector.InterferometerList
-        interferometers: ${data.detectors}
-      ref_time: ${data.ref_time}
-    - ${data.gnpe_time_shifts}
-    - _target_: dingo.gw.transforms.ProjectOntoDetectors
-      _partial_: true
-      _runtime_dependencies_: [domain]
-      ifo_list:
-        _target_: bilby.gw.detector.InterferometerList
-        interferometers: ${data.detectors}
-      ref_time: ${data.ref_time}
-    - _target_: dingo.gw.transforms.SampleNoiseASD
-      _partial_: true
-      _runtime_dependencies_: [asd_dataset]
-    - _target_: dingo.gw.transforms.WhitenAndScaleStrain
-      _partial_: true
-      _runtime_dependencies_: [domain]
-    - _target_: dingo.gw.transforms.AddWhiteNoiseComplex
-    - _target_: dingo.gw.transforms.SelectStandardizeRepackageParameters
-      parameters_dict:
-        inference_parameters: ${data.inference_parameters}
-        context_parameters: ${data.context_parameters}
-      standardization_dict: ${data.standardization}
-    - _target_: dingo.gw.transforms.RepackageStrainsAndASDS
-      _partial_: true
-      _runtime_dependencies_: [domain]
-      ifos: ${data.detectors}
-    - _target_: dingo.gw.transforms.UnpackDict
-      selected_keys: ${data.selected_keys}
-
-training:
-  stage_0:
-    epochs: 300
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: true
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      lr: 5.0e-5
-    scheduler:
-      T_max: 300
-  stage_1:
-    epochs: 150
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset/asds_O1.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: false
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      _target_: torch.optim.Adam
-      _partial_: true
-      lr: 1.0e-5
-    scheduler:
-      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
-      _partial_: true
-      T_max: 150
-
-local:
-  device: cuda
-  num_workers: 32
-  runtime_limits:
-    max_time_per_run: 3600000
-    max_epochs_per_run: 500
-  checkpoint_epochs: 10
-  leave_waveforms_on_disk: true
-  condor:
-    num_cpus: 16
-    memory_cpus: 128000
-    num_gpus: 1
-    memory_gpus: 8000
-    request_disk: 50GB
diff --git a/configs/experiment/train_gnpe_init.yaml b/configs/experiment/train_gnpe_init.yaml
deleted file mode 100644
index a56de527f..000000000
--- a/configs/experiment/train_gnpe_init.yaml
+++ /dev/null
@@ -1,70 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /model: gnpe_init
-  - _self_
-
-data:
-  waveform_dataset:
-    file_name: training_data/waveform_dataset.hdf5
-    domain_update:
-      f_min: 20.0
-      f_max: 1024.0
-    svd_size_update: 200
-  train_fraction: 0.95
-  detectors: [H1, L1]
-  ref_time: 1126259462.391
-  inference_parameters:
-    - H1_time
-    - L1_time
-
-training:
-  stage_0:
-    epochs: 150
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: true
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      lr: 5.0e-5
-    scheduler:
-      T_max: 150
-  stage_1:
-    epochs: 75
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset/asds_O1.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: false
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      _target_: torch.optim.Adam
-      _partial_: true
-      lr: 1.0e-5
-    scheduler:
-      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
-      _partial_: true
-      T_max: 75
-
-local:
-  device: cpu
-  num_workers: 6
-  runtime_limits:
-    max_time_per_run: 3600000
-    max_epochs_per_run: 500
-  checkpoint_epochs: 10
-  leave_waveforms_on_disk: true
-  condor:
-    num_cpus: 16
-    memory_cpus: 128000
-    num_gpus: 1
-    memory_gpus: 8000
-    request_disk: 50GB
diff --git a/configs/experiment/train_npe.yaml b/configs/experiment/train_npe.yaml
deleted file mode 100644
index f9413befd..000000000
--- a/configs/experiment/train_npe.yaml
+++ /dev/null
@@ -1,78 +0,0 @@
-# @package _global_
-
-defaults:
-  - override /model: npe
-  - _self_
-
-data:
-  waveform_dataset:
-    file_name: training_data/waveform_dataset.hdf5
-    domain_update:
-      f_min: 20.0
-      f_max: 1024.0
-    svd_size_update: 200
-  train_fraction: 0.95
-  detectors: [H1, L1]
-  ref_time: 1126259462.391
-  inference_parameters:
-    - chirp_mass
-    - mass_ratio
-    - chi_1
-    - chi_2
-    - theta_jn
-    - dec
-    - ra
-    - geocent_time
-    - luminosity_distance
-    - psi
-
-training:
-  stage_0:
-    epochs: 300
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset_fiducial/asds_O1_fiducial.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: true
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      lr: 5.0e-5
-    scheduler:
-      T_max: 300
-  stage_1:
-    epochs: 150
-    asd_dataset:
-      _target_: dingo.gw.noise.asd_dataset.ASDDataset
-      file_name: training_data/asd_dataset/asds_O1.hdf5
-      ifos: ${data.detectors}
-      precision: single
-      domain_update: null
-    freeze_rb_layer: false
-    batch_size: 4096
-    early_stopping: null
-    optimizer:
-      _target_: torch.optim.Adam
-      _partial_: true
-      lr: 1.0e-5
-    scheduler:
-      _target_: torch.optim.lr_scheduler.CosineAnnealingLR
-      _partial_: true
-      T_max: 150
-
-local:
-  device: cuda
-  num_workers: 32
-  runtime_limits:
-    max_time_per_run: 3600000
-    max_epochs_per_run: 500
-  checkpoint_epochs: 50
-  leave_waveforms_on_disk: true
-  condor:
-    num_cpus: 16
-    memory_cpus: 128000
-    num_gpus: 1
-    memory_gpus: 8000
-    request_disk: 50GB
diff --git a/configs/extrinsic_prior/default.yaml b/configs/extrinsic_prior/default.yaml
deleted file mode 100644
index 25b4a17c7..000000000
--- a/configs/extrinsic_prior/default.yaml
+++ /dev/null
@@ -1,8 +0,0 @@
-_target_: dingo.gw.prior.BBHExtrinsicPriorDict
-_convert_: all
-dictionary:
-  dec: bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')
-  ra: bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")
-  geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
-  psi: bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")
-  luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/extrinsic_prior/fmpe.yaml b/configs/extrinsic_prior/fmpe.yaml
deleted file mode 100644
index ab67e5e28..000000000
--- a/configs/extrinsic_prior/fmpe.yaml
+++ /dev/null
@@ -1,8 +0,0 @@
-_target_: dingo.gw.prior.BBHExtrinsicPriorDict
-_convert_: all
-dictionary:
-  dec: bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')
-  ra: bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")
-  geocent_time: bilby.core.prior.Uniform(minimum=-0.03, maximum=0.03, name='geocent_time')
-  psi: bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")
-  luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
diff --git a/configs/generate_asd_dataset.yaml b/configs/generate_asd_dataset.yaml
index aba442e62..6b128de6b 100644
--- a/configs/generate_asd_dataset.yaml
+++ b/configs/generate_asd_dataset.yaml
@@ -1,7 +1,3 @@
-defaults:
-  - hydra: default
-  - _self_
-
 data_dir: asd_dataset
 time_segments_file: null
 out_name: null
@@ -25,11 +21,19 @@ dataset_settings:
   observing_run: O1
 
 hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
   job:
     name: generate_asd_dataset
+    chdir: true
     config:
       override_dirname:
         exclude_keys:
           - data_dir
           - time_segments_file
           - out_name
+  job_logging:
+    loggers:
+      bilby:
+        level: INFO
+        propagate: true
diff --git a/configs/generate_dataset.yaml b/configs/generate_dataset.yaml
index 05838a47b..790134a1d 100644
--- a/configs/generate_dataset.yaml
+++ b/configs/generate_dataset.yaml
@@ -1,15 +1,36 @@
 defaults:
-  - hydra: default
   - domain: uniform_frequency
   - waveform_generator: phenom_d
-  - intrinsic_prior: default
   - _self_
 
+intrinsic_prior:
+  _target_: bilby.gw.prior.BBHPriorDict
+  _convert_: all
+  dictionary:
+    mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
+    mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
+    chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
+    mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
+    phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
+    chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
+    chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
+    theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
+    luminosity_distance: 100.0
+    geocent_time: 0.0
+
 num_samples: 10000
 compression: null
 num_processes: 1
 out_file: waveform_dataset.hdf5
 
 hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
   job:
     name: generate_dataset
+    chdir: true
+  job_logging:
+    loggers:
+      bilby:
+        level: INFO
+        propagate: true
diff --git a/configs/generate_synthetic_asd_dataset.yaml b/configs/generate_synthetic_asd_dataset.yaml
deleted file mode 100644
index 71070af80..000000000
--- a/configs/generate_synthetic_asd_dataset.yaml
+++ /dev/null
@@ -1,33 +0,0 @@
-defaults:
-  - hydra: default
-  - _self_
-
-asd_dataset: ???
-num_samples: 500
-num_processes: 1
-out_file: synthetic_asd_dataset.hdf5
-verbose: false
-
-parameterization_settings:
-  num_spline_positions: 30
-  num_spectral_segments: 400
-  sigma: 0.14
-  delta_f: -1
-  smoothen: true
-
-sampling_settings:
-  bandwidth_spectral: 0.5
-  bandwidth_spline: 0.25
-  split_frequencies:
-    - 30
-    - 100
-  rescaling_asd_paths: null
-
-hydra:
-  job:
-    name: generate_synthetic_asd_dataset
-    config:
-      override_dirname:
-        exclude_keys:
-          - asd_dataset
-          - out_file
diff --git a/configs/hydra/default.yaml b/configs/hydra/default.yaml
deleted file mode 100644
index f32382a96..000000000
--- a/configs/hydra/default.yaml
+++ /dev/null
@@ -1,11 +0,0 @@
-# @package _global_
-hydra:
-  run:
-    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
-  job:
-    chdir: true
-  job_logging:
-    loggers:
-      bilby:
-        level: INFO
-        propagate: true
diff --git a/configs/importance_weights.yaml b/configs/importance_weights.yaml
deleted file mode 100644
index 7f924ac7a..000000000
--- a/configs/importance_weights.yaml
+++ /dev/null
@@ -1,47 +0,0 @@
-defaults:
-  - hydra: default
-  - model: unconditional_npe
-  - optimizer: adam
-  - scheduler: cosine
-  - _self_
-
-outdir: .
-parameter_samples: ???
-num_samples: 1000
-event_dataset: null
-time_marginalization:
-  n_fft: 5
-phase_marginalization: null
-synthetic_phase: null
-calibration_marginalization: null
-num_processes: 1
-
-slice_plots:
-  num_slice_plots: 10
-  params_slice2d:
-    - [phase, geocent_time]
-    - [phase, tilt_1]
-
-nde:
-  path: null
-  data:
-    parameter_samples: null
-    parameters: null
-  model: ${model}
-  training:
-    optimizer: ${optimizer}
-    scheduler: ${scheduler}
-    device: cpu
-    num_workers: 0
-    train_fraction: 0.9
-    batch_size: 512
-    epochs: 20
-
-hydra:
-  job:
-    name: importance_weights
-    config:
-      override_dirname:
-        exclude_keys:
-          - parameter_samples
-          - nde.path
diff --git a/configs/intrinsic_prior/default.yaml b/configs/intrinsic_prior/default.yaml
deleted file mode 100644
index c9a697b87..000000000
--- a/configs/intrinsic_prior/default.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
-_target_: bilby.gw.prior.BBHPriorDict
-_convert_: all
-dictionary:
-  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
-  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
-  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
-  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
-  chi_1: bilby.gw.prior.AlignedSpin(name='chi_1', a_prior=Uniform(minimum=0, maximum=0.9))
-  chi_2: bilby.gw.prior.AlignedSpin(name='chi_2', a_prior=Uniform(minimum=0, maximum=0.9))
-  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
-  luminosity_distance: 100.0
-  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/fmpe.yaml b/configs/intrinsic_prior/fmpe.yaml
deleted file mode 100644
index 41d0f1737..000000000
--- a/configs/intrinsic_prior/fmpe.yaml
+++ /dev/null
@@ -1,17 +0,0 @@
-_target_: bilby.gw.prior.BBHPriorDict
-_convert_: all
-dictionary:
-  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
-  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
-  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=20.0, maximum=120.0, name='chirp_mass')
-  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
-  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
-  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
-  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
-  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
-  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
-  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
-  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
-  luminosity_distance: 100.0
-  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing.yaml b/configs/intrinsic_prior/precessing.yaml
deleted file mode 100644
index c2648da57..000000000
--- a/configs/intrinsic_prior/precessing.yaml
+++ /dev/null
@@ -1,17 +0,0 @@
-_target_: bilby.gw.prior.BBHPriorDict
-_convert_: all
-dictionary:
-  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_1')
-  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=120.0, name='mass_2')
-  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=150.0, name='chirp_mass')
-  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
-  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_1')
-  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.99, name='a_2')
-  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
-  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
-  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
-  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
-  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
-  luminosity_distance: 100.0
-  geocent_time: 0.0
diff --git a/configs/intrinsic_prior/precessing_multibanded_test.yaml b/configs/intrinsic_prior/precessing_multibanded_test.yaml
deleted file mode 100644
index 1db3d7c62..000000000
--- a/configs/intrinsic_prior/precessing_multibanded_test.yaml
+++ /dev/null
@@ -1,17 +0,0 @@
-_target_: bilby.gw.prior.BBHPriorDict
-_convert_: all
-dictionary:
-  mass_1: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_1')
-  mass_2: bilby.core.prior.Constraint(minimum=10.0, maximum=80.0, name='mass_2')
-  chirp_mass: bilby.gw.prior.UniformInComponentsChirpMass(minimum=15.0, maximum=100.0, name='chirp_mass')
-  mass_ratio: bilby.gw.prior.UniformInComponentsMassRatio(minimum=0.125, maximum=1.0, name='mass_ratio')
-  phase: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phase")
-  a_1: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_1')
-  a_2: bilby.core.prior.Uniform(minimum=0.0, maximum=0.88, name='a_2')
-  tilt_1: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_1')
-  tilt_2: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='tilt_2')
-  phi_12: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_12")
-  phi_jl: bilby.core.prior.Uniform(minimum=0.0, maximum=2*np.pi, boundary="periodic", name="phi_jl")
-  theta_jn: bilby.core.prior.Sine(minimum=0.0, maximum=np.pi, name='theta_jn')
-  luminosity_distance: 100.0
-  geocent_time: 0.0
diff --git a/configs/merge_asd_datasets.yaml b/configs/merge_asd_datasets.yaml
deleted file mode 100644
index 1dbafebb1..000000000
--- a/configs/merge_asd_datasets.yaml
+++ /dev/null
@@ -1,35 +0,0 @@
-defaults:
-  - hydra: default
-  - _self_
-
-data_dir: asd_dataset
-time_segments_file: null
-num_parts: -1
-out_name: null
-
-dataset_settings:
-  f_min: 0
-  f_max: 2048
-  f_s: 4096
-  time_psd: 1024
-  T: 8.0
-  window:
-    roll_off: 0.4
-    type: tukey
-  time_gap: 0
-  num_psds_max: 1
-  channels: null
-  detectors:
-    - H1
-    - L1
-  observing_run: O1
-
-hydra:
-  job:
-    name: merge_asd_datasets
-    config:
-      override_dirname:
-        exclude_keys:
-          - data_dir
-          - time_segments_file
-          - out_name
diff --git a/configs/model/unconditional_npe.yaml b/configs/model/unconditional_npe.yaml
deleted file mode 100644
index 3dc750652..000000000
--- a/configs/model/unconditional_npe.yaml
+++ /dev/null
@@ -1,12 +0,0 @@
-_target_: dingo.core.posterior_models.NormalizingFlowPosteriorModel
-embedding_kwargs: null
-posterior_kwargs:
-  num_flow_steps: 10
-  base_transform_kwargs:
-    hidden_dim: 128
-    num_transform_blocks: 2
-    activation: elu
-    dropout_probability: 0.1
-    batch_norm: true
-    num_bins: 8
-    base_transform_type: rq-coupling
diff --git a/configs/optimizer/adam.yaml b/configs/optimizer/adam.yaml
deleted file mode 100644
index 6182e57db..000000000
--- a/configs/optimizer/adam.yaml
+++ /dev/null
@@ -1,3 +0,0 @@
-_target_: torch.optim.Adam
-_partial_: true
-lr: 0.0001
diff --git a/configs/scheduler/cosine.yaml b/configs/scheduler/cosine.yaml
deleted file mode 100644
index 5ba12c5a7..000000000
--- a/configs/scheduler/cosine.yaml
+++ /dev/null
@@ -1,3 +0,0 @@
-_target_: torch.optim.lr_scheduler.CosineAnnealingLR
-_partial_: true
-T_max: 20
diff --git a/configs/train.yaml b/configs/train.yaml
index 8ef705247..f524d236d 100644
--- a/configs/train.yaml
+++ b/configs/train.yaml
@@ -1,15 +1,31 @@
 defaults:
-  - hydra: default
-  - extrinsic_prior: default
   - model: toy_npe
-  - optimizer: adam
-  - scheduler: cosine
   - _self_
 
 checkpoint: null
 train_dir: train
 exit_command: ""
 
+extrinsic_prior:
+  _target_: dingo.gw.prior.BBHExtrinsicPriorDict
+  _convert_: all
+  dictionary:
+    dec: bilby.core.prior.Cosine(minimum=-np.pi/2, maximum=np.pi/2, name='dec')
+    ra: bilby.core.prior.Uniform(minimum=0., maximum=2*np.pi, boundary="periodic", name="ra")
+    geocent_time: bilby.core.prior.Uniform(minimum=-0.10, maximum=0.10, name='geocent_time')
+    psi: bilby.core.prior.Uniform(minimum=0.0, maximum=np.pi, boundary="periodic", name="psi")
+    luminosity_distance: bilby.core.prior.Uniform(minimum=100.0, maximum=1000.0, name='luminosity_distance')
+
+optimizer:
+  _target_: torch.optim.Adam
+  _partial_: true
+  lr: 0.0001
+
+scheduler:
+  _target_: torch.optim.lr_scheduler.CosineAnnealingLR
+  _partial_: true
+  T_max: 20
+
 data:
   extrinsic_prior: ${extrinsic_prior}
   waveform_dataset:
@@ -115,5 +131,13 @@ local:
   test_only: false
 
 hydra:
+  run:
+    dir: ${oc.env:DINGO_RUNS}/${now:%Y-%m-%d_%H-%M-%S}_${hydra.job.name}_${hydra.job.override_dirname}
   job:
     name: train
+    chdir: true
+  job_logging:
+    loggers:
+      bilby:
+        level: INFO
+        propagate: true
diff --git a/configs/unconditional_density_estimation.yaml b/configs/unconditional_density_estimation.yaml
deleted file mode 100644
index 8129c0f5f..000000000
--- a/configs/unconditional_density_estimation.yaml
+++ /dev/null
@@ -1,29 +0,0 @@
-defaults:
-  - hydra: default
-  - model: unconditional_npe
-  - optimizer: adam
-  - scheduler: cosine
-  - _self_
-
-result_file: ???
-train_dir: nde
-
-data:
-  parameters: null
-
-training:
-  optimizer: ${optimizer}
-  scheduler: ${scheduler}
-  device: cpu
-  num_workers: 0
-  train_fraction: 0.9
-  batch_size: 512
-  epochs: 20
-
-hydra:
-  job:
-    name: unconditional_density_estimation
-    config:
-      override_dirname:
-        exclude_keys:
-          - result_file