Drop chains that are missing (structure) data in training #210
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Outdated suggestions cannot be applied.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.
This PR moves
chain_data_cache_paths
fromOpenFoldDataset
(where it was a list, one per dataset) to a single stringchain_data_cache_path
inOpenFoldSingleDataset
. This makes it convenient to drop entries (with a warning) where we have alignments but no structure data, instead of crashing.The motivation here was that while trying to start a small training run, I was hitting an issue where a few chains were present in my alignment data (downloaded from RODA) but absent in my pdb_mmcif dir (downloaded from PDB via download_pdb_mmcif.sh). I see there is now
scripts/download_roda_pdbs.sh
to address this but I thought it might be good to be able handle this error a bit better anyway.Curious if you think this makes sense @gahdritz
I also note a few tests are failing for me here, but the same tests appear to be failing for me on
main
(basic inference and training seems to work). If you approve the PR I'd rather have you wait though to merge it so I can check more carefully it's not breaking anything.