Complete dpctl tensor removal

[ethanglaser]

Description

Carrying out entirety of #2764

---

Checklist:

Completeness and readability

• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation to reflect the changes or created a separate PR with updates and provided its number in the description, if necessary.
• Git commit message contains an appropriate signed-off-by string (see CONTRIBUTING.md for details).
• I have resolved any merge conflicts that might occur with the base branch.

Testing

• I have run it locally and tested the changes extensively.
• All CI jobs are green or I have provided justification why they aren't.
• I have extended testing suite if new functionality was introduced in this PR.

[ethanglaser]

/intelci: run

[codecov]

Codecov Report

❌ Patch coverage is 82.35294% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
onedal/datatypes/_data_conversion.py	50.00%	0 Missing and 1 partial ⚠️
onedal/datatypes/_sycl_usm.py	80.00%	1 Missing ⚠️
onedal/utils/_array_api.py	0.00%	0 Missing and 1 partial ⚠️

Flag	Coverage Δ
azure	75.89% <82.35%> (?)
github	81.35% <82.35%> (-0.07%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
onedal/_device_offload.py	81.72% <100.00%> (ø)
onedal/datatypes/__init__.py	100.00% <100.00%> (ø)
onedal/datatypes/_dlpack.py	69.56% <ø> (ø)
onedal/datatypes/tests/common.py	84.31% <100.00%> (-0.88%)	⬇️
onedal/utils/_third_party.py	88.46% <ø> (-0.43%)	⬇️
sklearnex/_device_offload.py	77.41% <100.00%> (ø)
sklearnex/covariance/incremental_covariance.py	79.41% <ø> (ø)
sklearnex/decomposition/pca.py	92.67% <ø> (ø)
sklearnex/dummy/_dummy.py	82.55% <ø> (ø)
sklearnex/neighbors/_lof.py	100.00% <ø> (ø)
... and 6 more

... and 24 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ethanglaser]

/intelci: run

[ethanglaser]

CI Failure Triage Report

Run: 23814899308 | Date: 2026-03-31

Failed Jobs: LinuxNightly/venv Python3.14_Sklearn1.8

• LinuxNightly/venv Python3.14_Sklearn1.8 (failed step: Sklearn testing [preview])

Analysis

CI Failure Analysis

1. Failure Classification: PR-specific

2. Root Cause:

The failure is caused by the removal of the copy_to_usm function and related dpctl tensor support code. The test is likely hitting a code path where copy_to_usm is still being called but no longer exists.

3. Evidence:

• Similar failure pattern: The same test (test_ridge_regression_check_arguments_validity) failed on another branch with a TypeError: 'dict' object is not callable, suggesting this specific test is sensitive to array handling changes
• Incomplete log: The provided log cuts off abruptly in the middle of test execution, which often indicates a crash or unhandled exception
• Code removal impact: The PR removes significant functionality around dpctl tensor handling without updating all call sites

4. Relevant Code Changes:

The most problematic changes in the diff are:

# In onedal/_device_offload.py - Line 162-166
# BEFORE:
return (
    copy_to_dpnp(queue, result)
    if is_dpnp_ndarray(data)
    else copy_to_usm(queue, result)  # This function was removed!
)
# AFTER:
return copy_to_dpnp(queue, result)

And the complete removal of copy_to_usm from:

• onedal/datatypes/__init__.py (removed from exports)
• onedal/datatypes/_sycl_usm.py (function deleted)

5. Recommendation:

Immediate actions needed:

1. Find remaining references: Search the entire codebase for any remaining calls to copy_to_usm() that weren't updated in this PR
2. Update device offload logic: The simplified logic in _device_offload.py may be too aggressive - verify that copy_to_dpnp() can handle all the cases that copy_to_usm() previously handled
3. Test with dpctl arrays: Ensure the removal of dpctl tensor support doesn't break existing functionality that still relies on it during the deprecation period
4. Run local tests: Execute the failing test locally with the PR changes to reproduce and debug the specific error

Debugging steps:

# Run the specific failing test locally
pytest linear_model/tests/test_ridge.py::test_ridge_regression_check_arguments_validity -v

The issue likely stems from incomplete migration from the old copy_to_usm/copy_to_dpnp dual-path approach to the simplified dpnp-only approach.

---

Similar failures in recent runs

Job	Source	Date	Matching Errors
LinuxNightly/venv Python3.14_Sklearn1.8	dev/eglaser-dpctl-tensor-removal	2026-03-31	TypeError: 'dict' object is not callable, FAILED linear_model/tests/test_ridge.py::test_ridg

---

Generated by CI Triage Bot

[david-cortes-intel]

@ethanglaser You also need to remove here:

scikit-learn-intelex/sklearnex/linear_model/tests/test_logreg.py

Line 593 in 2079f27

- GPU: dpnp/dpctl arrays with newton-cg solver

[david-cortes-intel]

LGTM, pending CI results.

[ethanglaser]

/intelci: run

[Copilot]

Pull request overview

This PR completes the removal of dpctl.tensor support across sklearnex and onedal, aligning tests, utilities, and data conversion/offload paths to use dpnp (and/or Array API) instead.

Changes:

• Remove dpctl.tensor handling from array-namespace detection and table/data conversion utilities.
• Update GPU/SPMD and validation test matrices to stop parametrizing dpctl tensor inputs and focus on dpnp + Array API.
• Simplify SYCL output re-wrapping paths to return dpnp outputs for SYCL-USM inputs.

Reviewed changes

Copilot reviewed 42 out of 42 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
sklearnex/utils/tests/test_validation.py	Update supported dataframe list and comments to replace dpctl with dpnp.
sklearnex/utils/tests/test_class_weight.py	Remove dpctl from array API test parametrization.
sklearnex/utils/_array_api.py	Update comments/docstring wording to remove dpctl references.
sklearnex/tests/test_patching.py	Remove dpctl branches/skips and update comments for dpnp-only testing.
sklearnex/tests/test_common.py	Treat dpnp as an actively-loaded framework in lazy-import checks.
sklearnex/spmd/neighbors/tests/test_neighbors_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/linear_model/tests/test_logistic_regression_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/linear_model/tests/test_linear_regression_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/linear_model/tests/test_incremental_linear_spmd.py	Drop dpctl from GPU SPMD dataframe filters and update comments.
sklearnex/spmd/ensemble/tests/test_forest_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/decomposition/tests/test_pca_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/decomposition/tests/test_incremental_pca_spmd.py	Drop dpctl from GPU SPMD dataframe filters and update comments.
sklearnex/spmd/covariance/tests/test_incremental_covariance_spmd.py	Drop dpctl from GPU SPMD dataframe filters and update comments.
sklearnex/spmd/covariance/tests/test_covariance_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/cluster/tests/test_kmeans_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/cluster/tests/test_dbscan_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/spmd/basic_statistics/tests/test_incremental_basic_statistics_spmd.py	Drop dpctl from GPU SPMD dataframe filters and update comments.
sklearnex/spmd/basic_statistics/tests/test_basic_statistics_spmd.py	Drop dpctl from GPU SPMD dataframe filters.
sklearnex/preview/decomposition/tests/test_incremental_pca.py	Remove dpctl from dataframe parametrization.
sklearnex/preview/covariance/covariance.py	Update comments referring to dpnp/dpctl behavior to dpnp-only.
sklearnex/neighbors/knn_classification.py	Update comment examples to remove dpctl.
sklearnex/neighbors/common.py	Update comment about device-aware allocation to remove dpctl.
sklearnex/neighbors/_lof.py	Update comment about queue retention to remove dpctl.
sklearnex/manifold/tests/test_tsne.py	Update comment about numpy conversion to remove dpctl.tensor.
sklearnex/linear_model/tests/test_logreg.py	Update test docstring to remove dpctl mention.
sklearnex/dummy/tests/test_dummy.py	Remove dpctl from dataframe parametrization.
sklearnex/dummy/_dummy.py	Update module-level documentation comments to remove dpctl references.
sklearnex/decomposition/tests/test_pca.py	Remove dpctl from dataframe parametrization.
sklearnex/decomposition/pca.py	Update comment to remove dpctl mention.
sklearnex/covariance/incremental_covariance.py	Update comments referring to dpnp/dpctl behavior to dpnp-only.
sklearnex/_device_offload.py	Remove copy_to_usm usage and always rewrap SYCL outputs via copy_to_dpnp.
onedal/utils/tests/test_validation.py	Remove dpctl from dataframe parametrization in validation tests.
onedal/utils/_third_party.py	Remove is_dpctl_tensor helper and update dpctl-related comments.
onedal/utils/_array_api.py	Remove dpctl.tensor namespace mapping; SYCL namespace mapping now dpnp-only.
onedal/tests/utils/_dataframes_support.py	Remove dpctl dataframe support and default framework list entry.
onedal/datatypes/tests/test_data.py	Remove dpctl from dataframe parametrization; switch dlpack test to dpnp.
onedal/datatypes/tests/common.py	Update USM attribute assertion helper to operate on dpnp arrays only.
onedal/datatypes/_sycl_usm.py	Remove dpctl-based USM allocation path; implement device copy via dpnp.asarray.
onedal/datatypes/_dlpack.py	Update comment to remove dpctl reference.
onedal/datatypes/_data_conversion.py	Remove dpctl tensor return-type construction path and warnings.
onedal/datatypes/init.py	Stop exporting copy_to_usm; keep copy_to_dpnp.
onedal/_device_offload.py	Remove copy_to_usm usage and always rewrap SYCL outputs via copy_to_dpnp.