Describe the issue linked to the documentation

Hello,
if I'm not mistaken, I think that the documentation of RFECV about the cv parameter might be incorrect regarding the choice of StratifiedKFold or KFold when the estimator is a classifier. The docs read:

For integer/None inputs, if y is binary or multiclass, StratifiedKFold is used. If the estimator is a classifier or if y is neither binary nor multiclass, KFold is used.

Which matches the _rfe.py file.

I believe that the correct phrasing of the second sentence is that it's when the estimator is not a classifier, then KFold will be used. When a classifier is used (and y is binary or multiclass), it's possible to perform StratifiedKFold.

In fact, code-wise, the cv parameter is processed by check_cv, which states,

        For integer/None inputs, if classifier is True and ``y`` is either
        binary or multiclass, :class:`StratifiedKFold` is used. In all other
        cases, :class:`KFold` is used.

(here classifier is a boolean that is True when the estimator is a classifier).

According to the code of check_cv(), if cv is supplied as an integer (and so a cv method must be chosen), and the estimator is a classifier, and y is binary or multiclass, then the cv generator will indeed be a StratifiedKFold.

    cv = 5 if cv is None else cv
    if isinstance(cv, numbers.Integral):
        if (
            classifier
            and (y is not None)
            and (type_of_target(y, input_name="y") in ("binary", "multiclass"))
        ):
            return StratifiedKFold(cv)
        else:
            return KFold(cv)

Thank you!

Suggest a potential alternative/fix

I suggest changing the phrasing "If the estimator is a classifier" to "If the estimator is not a classifier".