Track SLEP10: Add n_features_in_ to all modules

This issue tracks the status of follow-ups of #18514, i.e. the implementation of SLEP10.

According to N_FEATURES_IN_AFTER_FIT_MODULES_TO_IGNORE in test_common.py, as of 96a96f19579c71da80a14f33a15a0402b2f797b4, modules to be n_feature_in_-ified are:

• calibration #19555
• compose #20175
• covariance #19341
• discriminant_analysis #19342
• ensemble #19326
• feature_extraction #20180
• feature_selection #19344
• isotonic #19539
• manifold #19539
• mixture #19540
• model_selection #20204
• multiclass #20193
• multioutput #19692
• naive_bayes #19485
• pipeline #20192
• random_projection #19541

Track documentation status of n_features_in_ , see N_FEATURES_MODULES_TO_IGNOREin test_docstring_parameters.py, start was #19351. Note that existing alternatives like n_features_ attributes have to be properly deprecated:

• calibration #19555
• cluster #20228
• cross_decomposition #19351
• compose #20175
• covariance #20236
• decomposition #20236
• discriminant_analysis #20236
• dummy #20236
• ensemble #20236
• feature_extraction
• feature_selection #20236
• gaussian_process #20236
• impute #20236
• isotonic
• kernel_approximation #20236
• kernel_ridge #20236
• linear_model #20236
• manifold #20236
• mixture #19540
• model_selection #20204
• multiclass #20193
• multioutput #19692
• naive_bayes #20236
• neighbors #20236
• neural_network #20236
• pipeline #20192
• preprocessing #20236
• random_projection
• semi_supervised #20236
• svm #20236
• tree #20236

Note: for meta-estimators it is better to delegate feature consistency validation to the inner base estimators.

[ogrisel]

Yeah!!!

[thomasjpfan]

How do we document n_features_in_? In the merged PRs, I don’t see entries as attribute in the docstrings.

I think we should add them to the docstring. I opened https://github.com/scikit-learn/scikit-learn/pull/19351 to start the process of documenting this.

Another question is how we want to deal the attribute n_features_ in BaggingClassifier. Do we want to deprecate that one or do we just replace it by n_features_in_?

I would say to deprecate it.