Preserve dtype for NumPy array-protocol inputs

[viclafargue]

Closes #8430.

Problem : The issue seems to be unrelated to the GB300 platform. Scikit-learn’s array-like compatibility checks wrap float64 data in an object without a direct dtype attribute. cuML consequently defaulted the wrapped input to float32, producing different results from the original float64 input.

Solution : Normalize objects implementing array before dtype selection to preserve their represented dtype, and remove the stale xfails.

[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: b6574e1a-f0e2-424d-889f-7dba1138f273

📥 Commits

Reviewing files that changed from the base of the PR and between bc14f49 and c7ab6d3.

📒 Files selected for processing (3)

• python/cuml/cuml/internals/validation.py
• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
• python/cuml/tests/test_validation.py

💤 Files with no reviewable changes (1)

• python/cuml/tests/test_validation.py

🚧 Files skipped from review as they are similar to previous changes (2)

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

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved array validation for array-like inputs that provide __array__, preserving their original data type.
  • Enhanced compatibility across supported memory types and allowed data types.
  • Updated compatibility reporting for estimator and dimensionality-reduction checks to reflect current behavior.

• Tests

  • Added coverage for dtype preservation across device, host, and unspecified memory.
  • Removed obsolete expected-failure markers for non-array input checks.

Walkthrough

check_array now infers dtypes from objects implementing __array__. Tests cover float32 and float64 inputs across memory types. The sklearn compatibility xfail entries and explanations now match current failures.

Changes

Array validation and compatibility

Layer / File(s)	Summary
Array protocol dtype inference python/cuml/cuml/internals/validation.py, python/cuml/tests/test_validation.py	check_array materializes array-protocol objects before dtype selection. Tests verify float32 and float64 preservation across supported memory types.
Non-array compatibility checks python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml, python/cuml/tests/test_sklearn_compatibility.py	The xfail list replaces four estimator entries with the RandomForestRegressor non-array check and removes its duplicate entry. Remaining explanations describe repeated-fit prediction differences and inconsistent fit_transform and transform outputs.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• rapidsai/cuml#8366: Updates overlapping sklearn compatibility xfail handling in the same test files.

Suggested reviewers: csadorf

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 40.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the primary change: preserving dtype for NumPy array-protocol inputs.
Description check	✅ Passed	The description explains the dtype bug, its impact on compatibility checks, and the implemented fix.
Linked Issues check	✅ Passed	The changes address issue #8430 by preserving wrapped input dtypes and updating related compatibility test expectations.
Out of Scope Changes check	✅ Passed	The code, tests, xfail updates, and SPDX attribution change remain related to the dtype compatibility fix.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[betatim]

This looks like a reasonable fix.

[jcrist]

Fix looks good to me! A bit disappointing that this didn't fix the issues in test_sklearn_compatibility.py though.

[viclafargue]

/merge