Actually it seems to break something. Let's validate properly:
self._validate_data(X, force_all_finite=False, dtype=None, ensure_2d=False, reset=True|False)
I tested locally and it seems enough

Actually it seems to break something. Let's validate properly: self._validate_data(X, force_all_finite=False, dtype=None, ensure_2d=False, reset=True|False) I tested locally and it seems enough

This can create a new numpy array when X is a list and _ConstantPredictor does not actually use the values of X.

Thinking through the test such as check_fit2d_predict1d with meta-estimators: we are not validating in the meta-estimator and delegating to the inner estimator. This means we can not expect check_fit2d_predict1d to pass for meta-estimators where the inner estimator does not validate.

On a related note, I have been seeing the CI failing which is related related to the issue when the inner estimator is not validating. In that case test_estimators[OutputCodeClassifier(estimator=LogisticRegression(C=1))-check_complex_data] is failing.

Thanks @thomasjpfan