running pg-benchmark does not take into account dataset splits. #114
Description
Summary
I realise we don't have the splitting functionality in the dataset yet, but should not forget to implement this in the benchmark when possible.
Current examples do not take a split dataset, they take the full dataset and set this to the supervise/zero-shot benchmark.
Motivation
To avoid confusion we should only utilize:
- train, validation, test datasets in supervised.
- test dataset in zero-shot.
Proposed solution
Depends on execution of ProteinGym/proteingym-base#294