two types of validation data for fine tuning experiment

[bhomass]

sciplex_ood_splits has two declarations for split_ood_finetuning = "test".

One looks up the 32 validation_drugs, the second one uses random splits from train.

what is idea behind the validate drugs? how were they selected?

Ultimately which is the better version to use?

[MxMstrmn]

HI @bhomass,

the selection was done according to the original paper and their analysis of perturbation effect. Could you clarify, what you mean by 'sciplex_ood_splits has two declarations for split_ood_finetuning = "test".' ?