FIX avoid race condition in RFECV #23560
Conversation
| @@ -721,7 +721,7 @@ def fit(self, X, y, groups=None): | |||
| func = delayed(_rfe_single_fit) | |||
|
|
|||
| scores = parallel( | |||
| func(rfe, self.estimator, X, y, train, test, scorer) | |||
| func(rfe, clone(self.estimator), X, y, train, test, scorer) | |||
There was a problem hiding this comment.
Can we construct a non-regression test for this?
There was a problem hiding this comment.
Hi @thomasjpfan, I tried to reproduce the error in the original issue but wasn't able to do so :(. Could you help take a look if you got time, thank you!
There was a problem hiding this comment.
I am not sure if this is where the original error comes from. func = _rfe_single_fit does not actually mutate estimator:
_rfe_single_fit only uses estimator for _safe_split and the lambda function has it's own locally scoped `estimator.
scikit-learn/sklearn/feature_selection/_rfe.py
Lines 40 to 42 in 2f787f4
There was a problem hiding this comment.
Shall I close this PR for now ? Since it seems not related to the original issue
There was a problem hiding this comment.
Yea, I am okay with closing.
Reference Issues/PRs
Fixes #23533.
What does this implement/fix? Explain your changes.
Clones estimator before passing into parallel scoring jobs in
RFECV.