Summary

The test_logistic_regression_weighting test is failing due to a coefficient shape mismatch between scikit-learn and cuML implementations for multiclass logistic regression with class weighting.

Failing test/component: test_logistic_regression_weighting[multiclass-3-class_weight]

Failure observed in:

• https://github.com/rapidsai/cuml/actions/runs/18186376814/job/51771408187#step:10:1787
• https://github.com/rapidsai/cuml/actions/runs/18404785497/job/52442084686#step:11:776

Environment

• OS: rockylinux8
• Python version: 3.12.11
• CUDA version: 13.0.1
• GPU: L4
• Dependencies: latest-deps

Test Details

• Test file: python/cuml/tests/test_linear_model.py
• Test name: test_logistic_regression_weighting[multiclass-3-class_weight]
• Error message:

  AssertionError: assert <array_equal: [[ 0.15458884  0.57219807  0.20086461  0.72615748  0.28467592] [-0.04120884 -0.95653517  0.27531644 -0.0... -0.0342683 ] [-0.02566717  0.85933513 -0.42952511 -0.25958989 -0.0948978 ]] unit_tol=0.2 total_tol=0.3 with_sign=True>
  

Additional output from test:

coef.shape:  (5, 3)
coef:
 [[ 0.06129447 -0.02292896  0.00468886]
 [ 0.48544116 -0.81974606  0.46844332]
 [-0.59215055 -0.28511068 -0.74525061]
 [ 0.58522586  0.33769497  0.29343246]
 [-0.25972732 -0.3635446  -0.37287772]]
cucoef.shape:  (3, 5)
cucoef:
 [[ 0.10089213  0.89434892 -0.08919029  0.39662385  0.15714772]
 [-0.0411478  -0.95494912  0.28194222 -0.07558644 -0.0342683 ]
 [-0.02566717  0.85933513 -0.42952511 -0.25958989 -0.0948978 ]]

Root Cause Analysis