Support for sample_weight in the fit method

[kota7]

Is there any plan or demand for supporting sample_weight option in the fit method of estimators?

Since a number of model classes in scikit-learn support it, I suppose it would be useful for estimators such as StackingRegressor (provided that all regressors accept it).

I saw a related PR #255 about the support for fit_params options. Once this is merged, sample_weight will be implementable by passing same weights to all regressors and meta-regressor. Even so, API like fit(X, y [, sample_weight]) can be handy.

[rasbt]

I think support for sample_weight would be a good thing to have. I think if the fit methods support fit_params the sample_weight would be supported automatically, right? I am currently not sure whether #255 is still active. If not, would you like to take it over?

[kota7]

@rasbt I will have a look. At first glance, it looks close but not entirely automatic. In #255 it looks fit_params are given per estimator basis, while support for sample_weight requires options applied to all estimators in the both levels.

[rasbt]

Good point. Maybe having a separate sample_weights param would be the better option then.