QN Test Failure with Softmax Loss Function

[csadorf]

The QN (Quasi-Newton) test is failing for the following configuration:

• Loss function: softmax
• L2 regularization
• Learning rate: 0.1
• Penalty strength: 0.0
• Float32 precision

The test asserts that the model achieves a minimum score of 0.50, but is only achieving 0.4904.

Failure Details

The test failure occurs in the test_qn function with the following parameters:

• Test case: test_qn[False-0.1-0.0-l2-float32-softmax]
• Python version: 3.11.12
• Environment: Linux

The specific assertion that failed is:

assert cuml_score >= 0.50
AssertionError: assert 0.4904 >= 0.5

The test is located in /__w/cuml/cuml/python/cuml/cuml/tests/test_qn.py on line 71.