creation of train/val split not filters for ids not present in test set instead of smiles not present in test set (data set should now be 2,000-3,000 entities larger)
(this was the cause of some classes not having 50 members across train/val/test sets)
ChebiOverXPartial now also includes top_class into dataset instead of just subclasses
requirements has been updated
loss is now a package
weighting for chebi50_v227 added
model_kwargs are passed to model in evaluate_model()
changed black version to 24.2.0 (same as used by lint)
Various minor fixes, including:
ChebiOverXPartialnow also includestop_classinto dataset instead of just subclassesrequirementshas been updatedlossis now a packagechebi50_v227addedmodel_kwargsare passed to model inevaluate_model()