When introducing the _DynamicDataset in #39, we changed the meaning of raw files. Previously, the labeled train.pkl, validation.pkl and test.pkl file names were returned by raw_file_names_dict() and raw_file_names(). Now, the GO class has a raw_file_names_dict() method that only the direct downloads (e.g., for GO, the go-basic.obo and uniprot_sprot.dat). It overwrites the method in _DynamicDataset which returns data.pkl. raw_file_names() is missing completely.
Solution
[x] Introduce a third file names property for data.pkl
[x] Link raw_file_names to raw_file_names_dict
[x] Add chebi.obo as raw file name for chebi classes
[ ] Use data.pkl in weighted BCE loss (for calculating weights) (@sfluegel05)
Problem
When introducing the
_DynamicDatasetin #39, we changed the meaning ofrawfiles. Previously, the labeledtrain.pkl,validation.pklandtest.pklfile names were returned byraw_file_names_dict()andraw_file_names(). Now, the GO class has araw_file_names_dict()method that only the direct downloads (e.g., for GO, thego-basic.oboanduniprot_sprot.dat). It overwrites the method in_DynamicDatasetwhich returnsdata.pkl.raw_file_names()is missing completely.Solution
data.pklraw_file_namestoraw_file_names_dictchebi.oboas raw file name for chebi classesdata.pklin weighted BCE loss (for calculating weights) (@sfluegel05)