ENH Array Api support for linear, polynomial and sigmoid kernels

[OmarManzoor]

Reference Issues/PRs

Towards #26024

What does this implement/fix? Explain your changes.

• Adds array api support for linear, polynomial and sigmoid kernels in sklearn.pairwise

Any other comments?

CC: @ogrisel @adrinjalali @betatim

Note: I did check that the CUDA tests seem to pass.

[github-actions]

✔️ Linting Passed

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

_{Generated for commit: 5122ff8. Link to the linter CI: here}

[ogrisel]

I checked the code for polynomial_kernel and linear_kernel and they indeed seem to be ok if left unchanged, so just adding them to the array API test suite should indeed be enough.

Assuming CI is still green after conflict resolution, LGTM.

[ogrisel]

The CUDA run was green:

https://github.com/scikit-learn/scikit-learn/actions/runs/10451046790/job/28936627899

but the status was removed from the list once I added the "waiting for reviewer" label.

[ogrisel]

There was a random 403 error calling fetch_california_housing on pylatest_conda_forge_mkl. Re-running.

[ogrisel]

@betatim this one should be quick to review and merge and would unblock the other linked PR.

[OmarManzoor]

@adrinjalali Could you kindly review this PR? I think it should be quick review.

[adrinjalali]

I'm confused, safe_sparse_dot seems to support array API, but at the end we have

https://github.com/OmarManzoor/scikit-learn/blob/5122ff8bbbe0f1d4a3ba7da62d73de69968f46c5/sklearn/utils/extmath.py#L212

But I can't find toarray in the array API standard. So what am I missing?

[OmarManzoor]

That part of the code only runs for sparse arrays. So it isn't linked to the array api.