[CI] test_spectral_clustering_output_shape_type_and_label_count fails on arm64 with oldest-deps

[csadorf]

Summary

Spectral clustering test failure due to convergence issue with specific parameter configuration discovered by Hypothesis property-based testing.

Failing test/component: test_spectral_clustering.py::test_spectral_clustering_output_shape_type_and_label_count

Failure observed in:

• https://github.com/rapidsai/cuml/actions/runs/21328714927/job/61390105080
• https://github.com/rapidsai/cuml/actions/runs/21858035191/job/63079564258#step:12:1661

Environment

• OS: ubuntu-22.04
• Python version: 3.10.19, 3.11
• CUDA version: 12.2.2
• GPU: NVIDIA L4
• Architecture: arm64 (aarch64)
• Dependencies: oldest-deps

Test Details

• Test file: python/cuml/tests/test_spectral_clustering.py

• Test name: test_spectral_clustering_output_shape_type_and_label_count

• Error message:

  RuntimeError: Spectral clustering failed to converge. Ensure the input data has clear clustering structure.
  
  Falsifying example: test_spectral_clustering_output_shape_type_and_label_count(
      n_samples=1769,
      n_features=16,
      n_clusters=8,
      n_neighbors=11,
      affinity='nearest_neighbors',
      dtype=dtype('<f4'),
  )
  

• Traceback location: cuml/cluster/spectral_clustering.pyx, line 253

[csadorf]

@aamijar Would you mind having a look at this, please?

[aamijar]

I was not able to reproduce the issue on an L4 instance or a GH200 instance with the following environment:

Environment

OS: ubuntu-24.04
Python version: 3.13.11
CUDA version: 12.9
GPU: NVIDIA L4
Architecture: x86_64
Dependencies: latest-deps
Branch: release/26.02

To test locally I added the hypothesis example the test case as follows:

@example(
    n_samples=1769,
    n_features=16,
    n_clusters=8,
    n_neighbors=11,
    affinity="nearest_neighbors",
    dtype=np.float32,
)

[csadorf]

Thanks for looking into this. Closing until we observe this again.

[csadorf]

Re-observed in https://github.com/rapidsai/cuml/actions/runs/21858035191/job/63079564258#step:12:1661 . The issue is still flaky.

[csadorf]

@aamijar I think we have to at least mark this test as xfailing while we root-cause the problem.

[aamijar]

Thanks for reporting, sure let's mark it as xfailing while we root-cause the problem.