FIX HistGradientBoosting raising ValueError with monotonic_cst and categorical_feature

[yuanx749]

Reference Issues/PRs

Fixes #28898

What does this implement/fix? Explain your changes.

HistGradientBoosting uses ColumnTransformer to preprocess the input X, which places the categorical features at the beginning of X. This PR adjusts the feature order in monotonic_cst correspondingly, to avoid the error when calling TreeGrower.

Any other comments?

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: ec90944. Link to the linter CI: here}

[ogrisel]

Thanks for the PR @yuanx749, the fix and the test look good. Can you please add an entry to the change log (doc/whats_new/1.5.rst)?

[yuanx749]

Thank you for reviewing @ogrisel. I added the entry to whats_new.

[ogrisel]

Actually, I think it would be worth making the test stronger by expanding test_predictions to insert some extra columns f_cat_a = rng.randint(low=0, high=9, size=n_samples) (and similarly f_b and f_c, before, between and after f_0 and f_1 and no monotonicity constraint on those columns (and no impact on y either).

This way we could better check the correctness of the monotonicity constraint remapping.

[ogrisel]

/cc @thomasjpfan.

[yuanx749]

Actually, I think it would be worth making the test stronger by expanding test_predictions to insert some extra columns f_cat_a = rng.randint(low=0, high=9, size=n_samples) (and similarly f_b and f_c, before, between and after f_0 and f_1 and no monotonicity constraint on those columns (and no impact on y either).

This way we could better check the correctness of the monotonicity constraint remapping.

Good point. I updated test_predictions, and all tests pass locally.

Then I think the additional non-regression test I added before can be removed.

[thomasjpfan]

Thanks for the PR @yuanx749 ! The fix and additional tests looks good to me.

[ogrisel]

Let's delete the redundant test but otherwise LGTM! Thanks for the fix.