check_estimator for estimators which require a mandatory extra fit param. #16756
Description
Right now the check_estimators will fail if the given estimator requires a parameter to fit other than X, y. fairlearn/fairlearn#342 is an example of where it fails.
In some use-cases, it makes sense to require an extra parameter. For instance, in the case of fairness, you may require the sensitive_features to be passed as well as the data itself. How should we handle these cases?
An alternative would be to have feature names during fit and set the sensitive feature names in __init__.
@thomasjpfan would you have an idea?