Fix: Array-API - avoid failing for numpy fit + predict with sparse or array-like X

[cakedev0]

Reference Issues/PRs

Fixes #34132

What does this implement/fix? Explain your changes.

• Add a condition in check_same_namespace to allow the cases:
  • [estimator xp is numpy, X is not an array]
  • [estimator is sparsified, X is a numpy array]
• Add a (hopefully) helpful error message for the case [estimator's xp is NOT numpy, X is not an array], for which the current message might be quite hard to understand.
• Add tests

AI usage disclosure

I used AI assistance for:

• Test/benchmark generation
• Research and understanding

[david-cortes-intel]

Thanks for looking into it.

But this would still error out in the cases where the attribute being checked is sparse. For example:

import os
os.environ["SCIPY_ARRAY_API"] = "1"

import numpy as np
rng = np.random.default_rng(seed=123)
X = rng.random(size=(20, 3))
y = rng.integers(2, size=X.shape[0])

from sklearn import set_config
set_config(array_api_dispatch=True)

from sklearn.linear_model import LogisticRegression
model = LogisticRegression().fit(X,y)
model.sparsify()
model.predict(X)

[cakedev0]

Indeed, good catch thanks.

Edit: Fixed.

[ogrisel]

@cakedev0 We also need a changelog entry for this fix.

[OmarManzoor]

A really minor error message content suggestion only. Otherwise LGTM.

Thank you @cakedev0