feat: implement benchmark field 2

[MicPie]

New PR based on PR #116 and issue #115 due to updated main branch.
PR #116 will be closed when this PR is finished.

[jackapbutler]

Hey @MicPie, for this use case of #108 will we continue to use split_col going forward but this denotes the split of the dataset not split in benchmarks?

[kjappelbaum]

Hey @MicPie, for this use case of #108 will we continue to use split_col going forward but this denotes the split of the dataset not split in benchmarks?

I think the meta.yaml are not supposed to say anything about the split we will use yet. The only thing we try to capture is what molecules have been part of a test set such that we might be able to also put them into our test set. Hence, I think this benchmark field is good enough for now (?)

[kjappelbaum]

@MicPie should I commit the fix for the 403 error directly to your PR?

[MicPie]

@MicPie should I commit the fix for the 403 error directly to your PR?

@kjappelbaum Yes, please, feel free to add to this branch.

[jackapbutler]

Hey @MicPie, for this use case of #108 will we continue to use split_col going forward but this denotes the split of the dataset not split in benchmarks?

I think the meta.yaml are not supposed to say anything about the split we will use yet. The only thing we try to capture is what molecules have been part of a test set such that we might be able to also put them into our test set. Hence, I think this benchmark field is good enough for now (?)

Ah cool, so if we find one with some existing train/val/test splits should we recombine it into one for uploading?

[MicPie]

Ah cool, so if we find one with some existing train/val/test splits should we recombine it into one for uploading?

@jackapbutler Yes, I would go for that. With that we can still filter it down later depending on our benchmarking setup.

[kjappelbaum]

If the checks pass now, it is OK to merge for me. Thanks! 💯