Ability to set a custom fold proportions for MultilabelStratifiedKFold (pass "r" to IterativeStratification)

[lopuhin]

For us it's useful to be able to set custom fold proportions when using MultilabelStratifiedKFold (essentially passing custom r to IterativeStratification). It's easy enough to extend outside of the lib (only _make_test_folds needs to be copied), but I wonder if such a feature could be useful in the library itself, what do you think?

And thanks for a great library!

[trent-b]

Thank you for the suggestion. Could you share a use case where it would be useful to set custom fold proportions (pass a custom r)? I am not opposed to the idea as long as there is a use case for it. If you would like, you could submit a Pull Request with the suggested change along with some test code that I could review and merge. Thanks!

[lopuhin]

Thank you! The use-case is that we want to distribute a part of the dataset across a fold subset (n-1 folds in our case), and then distribute the rest of the dataset across all folds - setting r on the second step allows to get folds of equal sizes. The reason why we want to do that is rather specific, that's why is could be quite a marginal use-case.

[trent-b]

Closing since there has not been additional activity.