feat: add SklearnTransformAdapter (wrap any scikit-learn transformer)

packages/openstef-models/src/openstef_models/transforms/general/__init__.py

@@ -20,6 +20,7 @@
 from openstef_models.transforms.general.scaler import Scaler
 from openstef_models.transforms.general.selector import Selector
 from openstef_models.transforms.general.shifter import Shifter
+from openstef_models.transforms.general.sklearn_adapter import SklearnTransformAdapter
 
 __all__ = [
     "OUTLIER_NAN_MASK_PREFIX",
@@ -34,4 +35,5 @@
     "Scaler",
     "Selector",
     "Shifter",
+    "SklearnTransformAdapter",
 ]

packages/openstef-models/src/openstef_models/transforms/general/sklearn_adapter.py

@@ -0,0 +1,118 @@
+# SPDX-FileCopyrightText: 2025 Contributors to the OpenSTEF project <openstef@lfenergy.org>
+#
+# SPDX-License-Identifier: MPL-2.0
+
+"""Adapter that exposes any scikit-learn transformer as a TimeSeriesTransform."""
+
+import importlib
+from typing import Any, override
+
+import pandas as pd
+from pydantic import Field, PrivateAttr
+
+from openstef_core.base_model import BaseConfig
+from openstef_core.datasets import TimeSeriesDataset
+from openstef_core.exceptions import NotFittedError
+from openstef_core.transforms import TimeSeriesTransform
+from openstef_models.utils.feature_selection import FeatureSelection
+
+
+class SklearnTransformAdapter(BaseConfig, TimeSeriesTransform):
+    """Adapt any scikit-learn transformer to the OpenSTEF ``TimeSeriesTransform`` interface.
+
+    The transformer is specified by its import path and constructor parameters rather than
+    as an object, so the configuration stays serializable (save/load-able). It is fitted on
+    the selected feature columns; its output then replaces those columns while the remaining
+    columns pass through unchanged. Output column names come from the transformer's
+    ``get_feature_names_out()``, so shape-changing transforms (e.g. PCA, one-hot encoders)
+    are handled the same way as shape-preserving ones (e.g. scalers).
+
+    ``features_added()`` is populated after ``fit()`` (the output names of some transformers
+    are only known once fitted).
+
+    Example:
+        >>> import pandas as pd
+        >>> from datetime import timedelta
+        >>> from openstef_core.datasets import TimeSeriesDataset
+        >>> from openstef_models.transforms.general import SklearnTransformAdapter
+        >>>
+        >>> data = pd.DataFrame(
+        ...     {"load": [100.0, 200.0, 300.0]},
+        ...     index=pd.date_range("2025-01-01", periods=3, freq="h"),
+        ... )
+        >>> dataset = TimeSeriesDataset(data, timedelta(hours=1))
+        >>> adapter = SklearnTransformAdapter(transformer_class="sklearn.preprocessing.StandardScaler")
+        >>> adapter.fit(dataset)
+        >>> transformed = adapter.transform(dataset)
+        >>> abs(float(transformed.data["load"].mean().round(6)))
+        0.0
+        >>> adapter.features_added()
+        []
+    """
+
+    transformer_class: str = Field(
+        description="Import path of the scikit-learn transformer, e.g. 'sklearn.preprocessing.StandardScaler'.",
+    )
+    transformer_params: dict[str, Any] = Field(
+        default_factory=dict,
+        description="Keyword arguments passed to the transformer's constructor.",
+    )
+    selection: FeatureSelection = Field(
+        default=FeatureSelection.ALL,
+        description="Features the transformer is applied to.",
+    )
+
+    _transformer: Any = PrivateAttr()
+    _is_fitted: bool = PrivateAttr(default=False)
+    _added_features: list[str] = PrivateAttr(default_factory=list)
+
+    @property
+    @override
+    def is_fitted(self) -> bool:
+        return self._is_fitted
+
+    @override
+    def model_post_init(self, context: Any) -> None:
+        module_path, _, class_name = self.transformer_class.rpartition(".")
+        if not module_path:
+            msg = f"transformer_class must be a fully qualified import path, got {self.transformer_class!r}."
+            raise ValueError(msg)
+        transformer_cls = getattr(importlib.import_module(module_path), class_name)
+        self._transformer = transformer_cls(**self.transformer_params)
+        if not hasattr(self._transformer, "get_feature_names_out"):
+            msg = f"{self.transformer_class} does not implement get_feature_names_out and cannot be adapted."
+            raise TypeError(msg)
+
+    @override
+    def fit(self, data: TimeSeriesDataset) -> None:
+        features = self.selection.resolve(data.feature_names)
+        self._transformer.fit(data.data[features])
+        output_names = list(self._transformer.get_feature_names_out(features))
+        self._added_features = [name for name in output_names if name not in data.feature_names]
+        self._is_fitted = True
+
+    @override
+    def transform(self, data: TimeSeriesDataset) -> TimeSeriesDataset:
+        if not self._is_fitted:
+            raise NotFittedError(self.__class__.__name__)
+
+        features = self.selection.resolve(data.feature_names)
+        output_names = list(self._transformer.get_feature_names_out(features))
+        transformed = pd.DataFrame(
+            self._transformer.transform(data.data[features]),
+            index=data.data.index,
+            columns=output_names,
+        )
+
+        # Replace the transformed inputs with the transformer's output, keep the rest.
+        passthrough = [column for column in data.data.columns if column not in features]
+        result = pd.concat([data.data[passthrough], transformed], axis=1)
+
+        return TimeSeriesDataset(data=result, sample_interval=data.sample_interval)
+
+    @override
+    def features_added(self) -> list[str]:
+        return self._added_features
+
+
+__all__ = ["SklearnTransformAdapter"]

packages/openstef-models/tests/unit/transforms/general/test_sklearn_adapter.py

@@ -0,0 +1,97 @@
+# SPDX-FileCopyrightText: 2025 Contributors to the OpenSTEF project <openstef@lfenergy.org>
+#
+# SPDX-License-Identifier: MPL-2.0
+
+from datetime import timedelta
+
+import pandas as pd
+import pytest
+
+from openstef_core.datasets import TimeSeriesDataset
+from openstef_core.exceptions import NotFittedError
+from openstef_models.transforms.general import SklearnTransformAdapter
+from openstef_models.utils.feature_selection import Include
+
+
+def _dataset() -> TimeSeriesDataset:
+    data = pd.DataFrame(
+        {"load": [100.0, 200.0, 300.0], "temperature": [20.0, 25.0, 30.0]},
+        index=pd.date_range("2025-01-01", periods=3, freq="h"),
+    )
+    return TimeSeriesDataset(data, timedelta(hours=1))
+
+
+def test_shape_preserving_transformer_scales_in_place():
+    """A scaler keeps the same columns and reports no added features."""
+    dataset = _dataset()
+    adapter = SklearnTransformAdapter(transformer_class="sklearn.preprocessing.StandardScaler")
+
+    adapter.fit(dataset)
+    result = adapter.transform(dataset)
+
+    assert set(result.data.columns) == {"load", "temperature"}
+    assert result.data["load"].mean() == pytest.approx(0.0, abs=1e-9)
+    assert result.data["load"].std(ddof=0) == pytest.approx(1.0)
+    assert adapter.features_added() == []
+
+
+def test_shape_changing_transformer_replaces_features_with_outputs():
+    """PCA replaces the input features with its components (the added features)."""
+    adapter = SklearnTransformAdapter(
+        transformer_class="sklearn.decomposition.PCA",
+        transformer_params={"n_components": 2, "random_state": 0},
+    )
+    dataset = _dataset()
+
+    adapter.fit(dataset)
+    result = adapter.transform(dataset)
+
+    added = adapter.features_added()
+    assert len(added) == 2
+    # the two input features are gone, replaced by the two components
+    assert set(result.data.columns) == set(added)
+    assert "load" not in result.data.columns
+
+
+def test_unselected_features_pass_through_unchanged():
+    """Only the selected features are transformed; the rest pass through untouched."""
+    dataset = _dataset()
+    adapter = SklearnTransformAdapter(
+        transformer_class="sklearn.preprocessing.StandardScaler",
+        selection=Include("load"),
+    )
+
+    adapter.fit(dataset)
+    result = adapter.transform(dataset)
+
+    assert result.data["temperature"].tolist() == [20.0, 25.0, 30.0]
+    assert result.data["load"].mean() == pytest.approx(0.0, abs=1e-9)
+
+
+def test_transform_before_fit_raises():
+    """transform() before fit() raises NotFittedError."""
+    adapter = SklearnTransformAdapter(transformer_class="sklearn.preprocessing.StandardScaler")
+
+    with pytest.raises(NotFittedError):
+        adapter.transform(_dataset())
+
+
+def test_config_round_trips_and_rebuilds_transformer():
+    """The config is serializable and reconstructs an equivalent transformer."""
+    adapter = SklearnTransformAdapter(
+        transformer_class="sklearn.decomposition.PCA",
+        transformer_params={"n_components": 3},
+    )
+
+    restored = SklearnTransformAdapter.model_validate(adapter.model_dump())
+
+    assert restored.transformer_class == "sklearn.decomposition.PCA"
+    assert restored.transformer_params == {"n_components": 3}
+    assert type(restored._transformer).__name__ == "PCA"
+    assert restored._transformer.n_components == 3
+
+
+def test_invalid_transformer_class_raises():
+    """An unimportable transformer_class fails fast at construction."""
+    with pytest.raises(ModuleNotFoundError):
+        SklearnTransformAdapter(transformer_class="not_a_real_module.Nope")