FEA Add RepeatedStratifiedGroupKFold as a new splitter #24227
Conversation
|
No worries, I can address the review @glemaitre |
|
I think for this it makes sense to have a |
|
@adrinjalali I think that we can add the class and the deprecation can come as a whole where we move into parameters instead of separate class. |
|
In general, it looks good. Just a couple of nitpicks. |
|
It seems silly to me to introduce a class which we already know we'd like to deprecate, while we can simply add an arg to |
| @@ -1242,6 +1253,39 @@ def test_repeated_stratified_kfold_determinstic_split(): | |||
| next(splits) | |||
|
|
|||
|
|
|||
| def test_repeated_stratified_group_kfold_determinstic_split(): | |||
There was a problem hiding this comment.
I think that we should have an additional test to be sure that we don't get the same split if we have different random state.
There is a typo for deterministic.
There was a problem hiding this comment.
Was this what you had in mind @glemaitre?
Fixed the typo (it was also misspelled in test_repeated_stratified_kfold_deterministic_split, so I fixed it there, too.
| X = [[1, 2], [3, 4], [5, 6], [7, 8], [9, 10]] | ||
| y = [1, 1, 1, 0, 0] | ||
| groups = [0, 0, 1, 1, 1] | ||
| random_state = 1944695409 |
There was a problem hiding this comment.
we could use 0 and not a fancy number.
* Bump versionadded * Fixed two typos of deterministic in tests * Added multiple random_state parameters to tests
|
Thanks for the feedback. Is there a PR or issue discussing deprecating the |
|
As for repeating, I agree that But generally, we also want to have a And generally, we can have a sinlge |
closes #24247
Reference Issues/PRs
This functionality was discussed in #13621.
What does this implement/fix? Explain your changes.
This adds a splitter class to
model_selectionthat repeatsStratifiedGroupKFoldn times.Any other comments?