Use cudf to_cupy dtype conversions

[csadorf]

Updates cuML cudf-to-cupy conversion paths to pass dtype directly through to_cupy, now that cuDF handles dtype coercion for numeric and non-numeric columns. Adds focused validation and encoder coverage for dtype-preserving CuPy output.

Follow-up to NVIDIA/cudf#22136

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: c3b253e3-065b-409f-8e06-e97b18aa4f23

📥 Commits

Reviewing files that changed from the base of the PR and between 661a320 and 2188f74.

📒 Files selected for processing (4)

• python/cuml/cuml/internals/validation.py
• python/cuml/cuml/preprocessing/encoders.py
• python/cuml/tests/test_ordinal_encoder.py
• python/cuml/tests/test_validation.py

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved dtype conversion handling for GPU array operations, ensuring requested data types are correctly applied during conversion.

• Tests

  • Added test coverage for dtype preservation in encoder GPU output operations.
  • Added validation tests for GPU array conversion type compatibility.

Walkthrough

Improved cudf-to-cupy dtype conversion by passing dtype directly to to_cupy() with fallback pre-casting. Applied changes to check_array validation logic and OrdinalEncoder output conversion, with parametrized tests validating dtype correctness in both cases.

Changes

Dtype Handling in cudf-to-cupy Conversions

Layer / File(s)	Summary
check_array cudf-to-cupy dtype fallback logic python/cuml/cuml/internals/validation.py, python/cuml/tests/test_validation.py	check_array dense cudf-to-device conversion now attempts to_cupy(dtype=...) directly; if unsupported dtype conversion raises TypeError, pre-casts the array and retries. Result wrapped with cp.asarray(..., dtype=..., order=...). Parametrized test validates correct dtype conversion from cudf.DataFrame to specified target dtype on device output.
OrdinalEncoder cupy output dtype conversion python/cuml/cuml/preprocessing/encoders.py, python/cuml/tests/test_ordinal_encoder.py	_get_output() passes na_value=np.nan and dtype=dtype directly to to_cupy(), replacing prior astype(dtype).to_cupy(...) pre-casting. Test validates OrdinalEncoder(output_type="cupy", dtype=np.float32) produces a cupy array with matching float32 dtype.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• rapidsai/cuml#8038: Related modifications to cudf conversion and dtype-coercion logic in cuml/internals/validation.py, including dtype fallback patterns and check_array refactoring.
• rapidsai/cuml#8052: Related updates to the cudf-to-cupy conversion path in cuml.internals.validation, including dtype and copy semantics handling.

Suggested labels

Cython / Python

Suggested reviewers

• betatim
• jcrist
• divyegala

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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