test(preprocessing): mark the inf-mask timing guards as slow

[aniruddhaadak80]

Issue

Fixes #1336

Motivation and Context

tests/test_preprocessing/test_inf_handling.py has two performance guards that assert a wall-clock ratio against the other code path:

# test__inf_mask__per_block_path_not_slower_than_pandas_on_fragmented
numpy_median = _median_runtime(lambda: _numpy_split_inf_masks(X))
pandas_median = _median_runtime(lambda: _inf_masks_pandas_only(X))

# ~1.5x faster locally; allow 25% slack so only a genuine regression fails.
assert numpy_median < pandas_median * 1.25, (...)

and its sibling test__inf_mask__pandas_path_not_slower_than_per_block_on_single_block, which asserts the same kind of ratio in the other direction.

Two problems:

1. The slow marker the docstring claims is not there. The first guard's docstring says it is "marked slow to run at merge time", but the only decorator on it is @pytest.mark.skipif(PANDAS_BELOW_3, ...). There is no @pytest.mark.slow, so pytest -m "not slow" in the PR matrix (.github/workflows/ci.yml) does not deselect it. The sibling has no marker and no such claim. The net effect is that two wall-clock assertions run in every pull request job.

2. The 1.25x budget does not survive shared-runner noise. The comparison is a ratio of two statistics.median values over 11 wall-clock runs of ~7-9ms of work. A busy shared runner is enough to invert it.

Observed failure

From the Test Package Compatibility (macos-15, 3.14, highest) job on a PR that does not touch clean.py or the inf-mask paths at all:

tests/test_preprocessing/test_inf_handling.py::test__inf_mask__per_block_path_not_slower_than_pandas_on_fragmented FAILED [ 54%]

E       AssertionError: per-block inf-mask path regressed: 8.79ms vs pandas 6.94ms

8.79 / 6.94 = 1.27x, just over the 1.25x budget. The same commit passes the other matrix legs, so the failure is a property of the runner rather than of the code. That makes unrelated PRs red and trains reviewers to discount the signal.

What this changes

This adds the marker the docstring already describes, and applies it to the sibling guard which has the same shape. Both now run only in the merge-time matrix, which is what the docstring intends and what a timing assertion is good for.

The deterministic checks are deliberately untouched and still run in every PR. Each guard asserts the two paths agree element-for-element before timing them:

np.testing.assert_array_equal(
    np.stack(_numpy_split_inf_masks(X)), np.stack(_inf_masks_pandas_only(X))
)

That assertion is the one that actually catches a regression in this code, and it is not wall-clock dependent, so it keeps its place in the PR gate.

---

Public API Changes

• No Public API changes
• Yes, Public API changes (Details below)

---

How Has This Been Tested?

Both directions of the marker were verified with pytest --collect-only, so the evidence is about test selection rather than about wall-clock numbers, which are not reproducible on an arbitrary machine.

Before the change, the guard is collected when the PR matrix does not filter it, confirming it is a PR-gate test today:

$ pytest tests/test_preprocessing/test_inf_handling.py -k per_block_path_not_slower --collect-only -q
1/52 tests collected (51 deselected)

With the first marker added, it is deselected as intended:

$ pytest -m "not slow" tests/test_preprocessing/test_inf_handling.py --collect-only -q
51/52 tests collected (1 deselected)

With both markers added, the two timing guards are deselected and every other test in the file still runs:

$ pytest -m "not slow" tests/test_preprocessing/test_inf_handling.py --collect-only -q
50/52 tests collected (2 deselected)

Lint and format, both clean:

$ ruff check tests/test_preprocessing/test_inf_handling.py
All checks passed!

$ ruff format --check tests/test_preprocessing/test_inf_handling.py
1 file already formatted

No changelog fragment is included: a maintainer applied the "no changelog needed" label, which the repo README lists as the right call for test refactoring that is not a bug fix. That label only takes effect once the fragment is gone, because action-towncrier-changelog fails when both are present.

---

Checklist

• The changes have been tested locally.
• Documentation has been updated (if the public API or usage changes).
• A changelog entry has been added (see changelog/README.md), or "no changelog needed" label requested.
• The code follows the project's style guidelines.
• I have considered the impact of these changes on the public API.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 9ab9f63. Configure here.}

[adrian-prior]

Thanks for setting this up! Can you remove the changelog, then I am happy to approve the PR!