Skip to content

[CI] Nightly scikit-learn tree node dtype breaks RandomForest and IsolationForest conversions #8537

Description

@csadorf

Summary

Nightly-dependency wheel tests report 19 tree reconstruction failures while converting cuML RandomForest and IsolationForest models to scikit-learn estimators. The same node dtype mismatch appears in 15 main tests and 4 cuml.accel integration or ONNX tests.

Failing test/component: RandomForest and IsolationForest scikit-learn conversion

Failure observed in:

Environment

  • OS: Ubuntu 26.04, amd64
  • Python version: 3.14.7
  • CUDA version: 13.3.0
  • GPU: RTX PRO 6000
  • Dependencies: nightly-deps; scikit-learn==1.10.dev0, treelite==4.7.1, cuml-cu13==26.10.0a88.post260831062823

Test Details

  • Test files:
    • python/cuml/cuml/tests/test_sklearn_import_export.py (5 failures)
    • python/cuml/cuml/tests/test_isolation_forest.py (10 failures)
    • python/cuml/cuml_accel_tests/integration/test_rf_classifier.py (1 failure)
    • python/cuml/cuml_accel_tests/integration/test_rf_regressor.py (1 failure)
    • python/cuml/cuml_accel_tests/test_onnx.py (2 failures)
  • Representative tests:
    • test_random_forest_classifier[False]
    • test_as_sklearn_scoring_parity[params0]
    • test_oob_score
    • test_onnx_classifier[RandomForestClassifier]
  • Error message:
    ValueError: node array from the pickle has an incompatible dtype:
    - expected: {'names': ['left_child', 'right_child', 'feature', 'threshold', 'left_cat_bitset', 'impurity', 'n_node_samples', 'weighted_n_node_samples', 'missing_go_to_left', 'split_kind'], 'formats': ['<i8', '<i8', '<i8', '<f8', ('<u4', (8,)), '<f8', '<i8', '<f8', 'u1', 'i1'], 'offsets': [0, 8, 16, 24, 32, 64, 72, 80, 88, 89], 'itemsize': 96, 'aligned': True}
    - got     : {'names': ['left_child', 'right_child', 'feature', 'threshold', 'left_cat_bitset', 'impurity', 'n_node_samples', 'weighted_n_node_samples', 'missing_go_to_left'], 'formats': ['<i8', '<i8', '<i8', '<f8', ('<u4', (8,)), '<f8', '<i8', '<f8', 'u1'], 'offsets': [0, 8, 16, 24, 32, 64, 72, 80, 88], 'itemsize': 96, 'aligned': True}
    

Root Cause Analysis

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingci

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions