Add a optional fit_param to enable custom MultiOutput fit process

[huangk10]

Reference Issues/PRs

See also #15953

What does this implement/fix? Explain your changes.

This allows the MultiOutputRegressor.fit to accept a **fit_params to custom estimator fit process.

Example:

class GradientBoostingRegressorWithFitParam(GradientBoostingRegressor):
    def fit(self, X, y, sample_weight=None, monitor=None, **fit_params):
        self.fit_params = fit_params
        super().fit(X, y, sample_weight=sample_weight)
        
X, y = datasets.make_regression(n_targets=3)
X_train, y_train = X[:50], y[:50]
X_test, y_test = X[50:], y[50:]

rgr = MultiOutputRegressor(GradientBoostingRegressorWithFitParam())
rgr.fit(X_train, y_train, customed_param=True)
y_pred = rgr.predict(X_test)

Any other comments?

To custom some of the estimators' fit processes, we have to redifine the estimators.fit functions with customed paramters besides sample_weight. Beacause the fit methods of most defined estimators in the sklearn package just have a sample_weight parameter.

[glemaitre]

Be aware that we don't want to have any tight with LightGBM or some of their parameters. It might make sense that MultiOutputRegressor accepts a **fit_params but it should be decorrelated with LightGBM when coding in scikit-learn.

[glemaitre]

A couple of changes. We also need a what's new entry.

[glemaitre]

If you could just parametrize the test as below. Otherwise, LGTM

[glemaitre]

LGTM. ping @thomasjpfan

[thomasjpfan]

Thank you @huangk10 !