In #15555 we removed the inconsistent pos_label attribute.

It would be good to allow users to specify which class is the semantically positive class and slice predict_proba or decision_function accordingly.

I'm not sure if that parameter should be called pos_label as it is somewhat semantically different. Maybe positive_label? positive_class? but it's not clear that that's different. We could do use_as_positive= but that's a bit obscure maybe?

Also see #15405 (comment)