A few sklearn compat test fixes

[jcrist]

A few tiny sklearn compat test fixups.

• Update xfail reasons to be correct for all currently xfailed tests
• Fix min samples/features for PCA/TruncatedSVD/TSNE
• Tag KNeighborsRegressor as multi-output
• Small fix for StandardScaler logic

At this point most of our remaining xfails are statistical issues with sample weights (sklearn has these too for several estimators, getting these correct is an ongoing process).

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: cb42f04b-ce3d-4600-a6c8-fdc706c62528

📥 Commits

Reviewing files that changed from the base of the PR and between ddabacc and 5488e64.

📒 Files selected for processing (1)

• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml

💤 Files with no reviewable changes (1)

• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Strengthened input validation for PCA, TruncatedSVD, and t-SNE (rejects datasets with fewer than two samples or features).
  • Improved StandardScaler partial-fit sample-count handling for per-feature tracking.
  • Updated KNeighborsRegressor metadata to correctly advertise multi-output support.
• Tests
  • Refreshed sklearn compatibility/estimator-check expectations, including updated xfail reasons and adjusted coverage for several estimators.

Walkthrough

The changes update sklearn-compatible input validation, StandardScaler sample-count handling, KNeighborsRegressor metadata, and estimator compatibility expectations across PCA, TruncatedSVD, TSNE, UMAP, ensemble, linear, neighbor, and preprocessing checks.

Changes

Scikit-learn compatibility

Layer / File(s)	Summary
Input validation and scaler state handling python/cuml/cuml/_thirdparty/sklearn/preprocessing/_data.py, python/cuml/cuml/decomposition/pca.pyx, python/cuml/cuml/decomposition/tsvd.pyx, python/cuml/cuml/manifold/t_sne.pyx, python/cuml/tests/test_sklearn_compatibility.py	StandardScaler uses np.isscalar and np.full for per-feature sample counts. PCA, TruncatedSVD, and TSNE require at least two samples and two features; TSNE’s corresponding one-sample and one-feature xfails are removed.
Estimator metadata and upstream checks python/cuml/cuml/neighbors/kneighbors_regressor.pyx, python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml	KNeighborsRegressor reports multi-output support. Upstream xfails remove PCA, TruncatedSVD, and estimator-cloning entries while retaining RandomForest classifier and regressor checks.
Compatibility expectation updates python/cuml/tests/test_sklearn_compatibility.py	Xfail reasons and expected checks are updated for sample-weight equivalence, dtype tolerance, KNeighborsRegressor multi-output behavior, UMAP, Lasso, ElasticNet, and StandardScaler.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Possibly related PRs

• rapidsai/cuml#8166: Updates overlapping sklearn compatibility xfail expectations for estimators including RandomForest, KNeighbors, and TSNE.

Suggested labels: sklearn-api-compat

Suggested reviewers: csadorf

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Title check	❓ Inconclusive	The title is related, but it's generic and doesn't convey the main changes beyond broad sklearn compatibility fixes.	Mention the concrete changes, such as PCA/TSVD/TSNE validation, KNeighborsRegressor multi-output tags, and StandardScaler/test xfail updates.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description check	✅ Passed	The description matches the diff: it mentions xfail cleanup, validation fixes, KNeighborsRegressor tagging, and StandardScaler logic.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 5

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@python/cuml/cuml/_thirdparty/sklearn/preprocessing/_data.py`:
- Around line 774-777: The scalar expansion in StandardScaler.partial_fit()
currently coerces fractional n_samples_seen_ values to int64. Preserve the
original scalar’s dtype when creating the broadcast array, and add an interop
regression test covering a migrated weighted StandardScaler resumed via
partial_fit().

In `@python/cuml/cuml/decomposition/pca.pyx`:
- Around line 482-483: PCA input validation currently rejects valid
single-feature datasets through ensure_min_features=2. Update the validation
configuration in the PCA fitting path to allow one feature, while retaining the
minimum sample requirement, so PCA(n_components=1) accepts shape (n_samples, 1)
input.

In `@python/cuml/cuml/decomposition/tsvd.pyx`:
- Around line 316-317: The TruncatedSVD validation currently rejects valid
one-row inputs via ensure_min_samples=2. In the TruncatedSVD input validation
logic, remove the sample minimum while retaining ensure_min_features=2, unless
the solver explicitly requires otherwise; add a regression test covering
fit_transform on a single-sample matrix.

In `@python/cuml/cuml/manifold/t_sne.pyx`:
- Around line 598-599: The TSNE input validation currently rejects valid
single-feature data via ensure_min_features=2. In the TSNE validation call near
ensure_min_samples, remove the ensure_min_features restriction so
one-dimensional inputs are accepted, and add or update coverage for a 1-feature
TSNE input.

In `@python/cuml/tests/test_sklearn_compatibility.py`:
- Around line 212-215: Restrict the KNeighborsRegressor exception for
check_regressor_multioutput to the float32-versus-float64 dtype assertion only,
rather than xfail-ing the entire check. Preserve or add an explicit assertion
for multi-output prediction shape so shape regressions cannot be hidden.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7e0e4bc3-5a1e-4db8-8fc0-635272809379

📥 Commits

Reviewing files that changed from the base of the PR and between f19ca67 and ddabacc.

📒 Files selected for processing (7)

• python/cuml/cuml/_thirdparty/sklearn/preprocessing/_data.py
• python/cuml/cuml/decomposition/pca.pyx
• python/cuml/cuml/decomposition/tsvd.pyx
• python/cuml/cuml/manifold/t_sne.pyx
• python/cuml/cuml/neighbors/kneighbors_regressor.pyx
• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
• python/cuml/tests/test_sklearn_compatibility.py

💤 Files with no reviewable changes (1)

• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml

[csadorf]

LGTM!

[jcrist]

/merge