feat(naive_bayes): dataframe-agnostic MultinomialNB via narwhals - #1974

Merging this PR will not alter performance

✅ 98 untouched benchmarks
⏩ 16 skipped benchmarks¹

_{Comparing JokeGbenro:feat/multinomial-nb-narwhals (a52e293) with main (fe248b4)}

Hi @MaxHalford @smastelini

I pushed two follow-up fixes for the failing checks.

The pandas doctest was failing because BaseNB.predict_many forced predicted labels into an object array, which changed GaussianNB.predict_many output from dtype: int64 to dtype: object. I removed the forced object dtype so numeric labels keep their native dtype.

The prek failure was from mypy in MultinomialNB.learn_many, where y was reused for both the input series and the one-hot sparse target matrix. I split that into y_one_hot and added the return annotation for _one_hot_targets.

Verified locally:

• python -m pytest river/naive_bayes -q
• python -m ruff check river\naive_bayes\multinomial.py
• targeted mypy check on river\naive_bayes\multinomial.py

Could you fix the conflicts?

Hi @MaxHalford, I’ve resolved the merge conflicts.

The Naive Bayes tests have been kept in their new tests/naive_bayes location, and the fixture import was updated to use tests.frames. I also verified that there are no remaining conflict markers and that the changed Python files compile successfully.

Ok good! Can you add a release note to unreleased.md?

Hi @MaxHalford, added the release note in unreleased.md under naive_bayes