FIX ensure consistency or column and feature names in FunctionTransformer

[glemaitre]

closes #27695

Raise an explicit error when the column names of the container given by transform is not consistent with the output of get_feature_names_out in FunctionTransformer.

In #27695, the error raised is not easy to understand when the FunctionTransformer is embedded within a Pipeline.

Here, we also give some solution how to resolve the problem.

I see that we have test failing in our test suite. I need to check if they are legitimate. I see that some come from the fact that feature_names_out return less names than the number of columns in X_trans.

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: 468d0b3. Link to the linter CI: here}

[glemaitre]

Calling transform allows us to find the issue with the number of columns and the function used here. Therefore, we can call fit to avoid this check.

[glemaitre]

So here, we only raise a better error message. There is no magic but we provide an explanation what to do.

I am not a big fan of the magical solution and I am not sure that we will be able to somehow return the expected type (NumPy vs. Pandas) since it will depend of what func is supposed to output (in the case that validate=False).

[thomasjpfan]

LGTM

[lorentzenchr]

LGTM

[lorentzenchr]

Needs sync with main before merging.

[glemaitre]

Thanks @lesteve for syncing the PR. Merging with the 2 above approvals.