ENH Add Feature Importances to _MultiOutputEstimator for Both Classifier and Regressor

[bewygs]

This implement feature_importances_ attribute in _MultiOutputEstimator when the base estimator supports it.

Reference Issues/PRs

To my knowledge, there are no open issues that this directly addresses or closes.

What does this implement/fix? Explain your changes.

This PR adds a feature_importances_ attribute to the _MultiOutputEstimator class to accommodate both classifiers and regressors from sklearn.ensemble.

Any other comments?

I believe it would be more convenient to be able to access feature_importances_ directly from our _MultiOutputEstimator object, rather than through a loop each time, as illustrated below :

feature_importances = [estimator.feature_importances_ for estimator in self.estimators_]

Example Usage:

import numpy as np
from sklearn.multioutput import MultiOutputRegressor, MultiOutputClassifier
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor

X, y = np.random.rand(100, 3), np.random.rand(100, 2)

# For regressor
rf_regressor = RandomForestRegressor(n_estimators=10)
multi_rf_regressor = MultiOutputRegressor(rf_regressor)
multi_rf_regressor.fit(X, y)
print("Average Feature Importances for Regressor:", multi_rf_regressor.feature_importances_)

# For classifier
y_classifier = np.random.randint(0, 2, size=(100, 2))
rf_classifier = RandomForestClassifier(n_estimators=10)
multi_rf_classifier = MultiOutputClassifier(rf_classifier)
multi_rf_classifier.fit(X, y_classifier)
print("Average Feature Importances for Classifier:", multi_rf_classifier.feature_importances_)

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 42860be. Link to the linter CI: here}

[adrinjalali]

In so many cases people shouldn't be using feature_importances_ anyway, and with all the discussions around this, I don't think we should add this.

cc @glemaitre

[glemaitre]

Indeed, I think we should instead invest time in scikit-learn/enhancement_proposals#86. Once we settle on this, then we can decide to implement different feature importances that can be chosen by the user.