Repository navigation
Conversation
Bump actions/checkout v3->v7, actions/cache v2.1.7->v6, and conda-incubator/setup-miniconda v2->v4. Replace the retired Mambaforge installer with Miniforge (which bundles mamba). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Un-nest pull_request trigger (was a child of push, so PRs never ran tests) - Use matrix.platform for runs-on so macOS/Windows jobs actually run - Run steps in a bash login shell so the conda env's Python is used on all platforms - Key cache on platform + Python version (matrix.os was undefined) - Rename job buld_cpu -> build_cpu Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Disable fail-fast in CI test matrix Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Fix CI install and test failures - Replace setup-miniconda with setup-python. Under a conda login shell poetry installed into its own environment and broke itself when the lockfile downgraded virtualenv (FileNotFoundError installing pytdc). - Test on Linux only; README states macOS/Windows are unsupported and the lockfile pins torch 2.1.2+cpu, which has no macOS arm64 wheel. - Switch the dummy dataset to PDB format: mmtf.rcsb.org has been retired and no longer resolves, failing test_train_fast_dev_run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Require Pillow >=13.0.0 (raises minimum Python to 3.11) (#112) * Require Pillow >=13.0.0 and Python 3.11 Pillow 13 requires Python >=3.11, so raise the supported range from >=3.9,<3.11 to >=3.11,<3.12 (torch 2.1.2 supports up to 3.11). Re-lock with existing pins kept: pillow 10.2.0 -> 13.0.0, and drop the async-timeout/exceptiongroup/importlib-resources backports. Update classifiers, ruff/black targets, README and the CI matrix to match. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Remove torchdrug dependency from GearNet torchdrug's last release (0.2.1, 2023) requires Python <3.11, so it cannot be installed alongside Pillow 13. GearNet only used it for utils.sparse_coo_tensor, an unchecked wrapper around torch.sparse_coo_tensor, and data.Graph.from_tensors inside _message_and_aggregate, which nothing calls. Use torch directly, drop the dead method, and remove the install-torchdrug task and CI step. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
a-r-j
changed the title
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Brings the follow-ups to #110 into
main: the CI fixes from #111 and the Pillow upgrade from #112. Both were merged into this branch after #110 was squash-merged.Warning
Breaking: drops Python 3.9 and 3.10 (Python 3.11 only), and removes
torchdrugalong with theinstall-torchdrugpoe task.CI fixes (#111)
actions/setup-python. In a conda login shell, poetry installed into its own environment and broke itself when the lockfile downgradedvirtualenv(FileNotFoundErrorinstallingpytdc). Poetry now uses its in-project.venv, which is the path the cache step saves.torch 2.1.2+cpuhas no macOS arm64 wheel.dummytest dataset now uses PDB format. RCSB retiredmmtf.rcsb.org, which broketest_train_fast_dev_run.fail-fast: false, so one Python version failing doesn't cancel the others.Pillow >=13.0.0 (#112)
pythonmoves from>=3.9,<3.11to>=3.11,<3.12. The upper bound comes from torch 2.1.2, which only supports up to 3.11.pillow10.2.0 → 13.0.0. The re-lock also dropsasync-timeout,exceptiongroupandimportlib-resources, which were only needed on Python <3.11.target-version, the README conda example and the CI matrix (["3.11"]).torchdrugremovedIts last release (0.2.1, 2023) requires Python <3.11. GearNet was the only user:
utils.sparse_coo_tensor→torch.sparse_coo_tensor. The torchdrug version was an unchecked wrapper around it, and I checked locally on torch 2.1.2 that the results match a dense reference.MessagePassingBase._message_and_aggregate, which nothing called and which was the only use ofdata.Graph.from_tensors.Testing
CI ✅
Build CPU (3.11): 20 passed, 6 skipped. ✅changelog.The GearNet forward-pass test is skipped in CI ("too slow for GitHub Actions"), so it's worth running GearNet locally before merging.
Follow-up
14 other dataset configs (CATH, ccPDB, antibody developability, …) still use
format: "mmtf". Fresh downloads for those will fail until they're switched.🤖 Generated with Claude Code