Description

Hi all,

In cross_val_predict, from the documentation it seems we can input groups and cv= integer/None to have both GroupKFold and StratifiedKFold constraints. However, I looked into the source code and I think there is no such K-fold (with both 2 constraints). In fact, as mentioned in the description the cross_val_predict use StratifiedKFold or KFold and in the cases groups are "Always ignored". So I was wondering if the groups in cross_val_predict cannot be used at all cases? (not sure in the case if we input GroupKFold() as the customized cv)

Also, if not implemented, is it possible to have a K-Fold that can have both constraints (stratified-group-k-fold) ? It might not always have a solution with 2 constraints satisfied, but I guess an approximated solution might be available.

Thanks!