Improve detection of scikit-learn parity regressions

ci/test_python_scikit_learn_tests.sh

@@ -2,7 +2,7 @@
 # Copyright (c) 2025, NVIDIA CORPORATION.
 
 # Support invoking test script outside the script directory
-cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../
+cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../ || exit 1
 
 # Common setup steps shared by Python test jobs
 source ./ci/test_python_common.sh
@@ -15,12 +15,14 @@ set +e
 rapids-logger "Running scikit-learn tests with cuML acceleration"
 
 # Run the tests
-./ci/accel/scikit-learn-tests/run-tests.sh \
-    --junitxml="${RAPIDS_TESTS_DIR}/junit-cuml-accel-scikit-learn.xml" || true
+./python/cuml/cuml/accel/tests/scikit-learn/run-tests.sh \
+    --numprocesses=8 \
+    --dist=worksteal \
+    --junitxml="${RAPIDS_TESTS_DIR}/junit-cuml-accel-scikit-learn.xml"
 
 # Analyze results and check pass rate threshold
 rapids-logger "Analyzing test results"
-./ci/accel/scikit-learn-tests/summarize-results.sh \
+./python/cuml/cuml/accel/tests/scikit-learn/summarize-results.py \
     --fail-below 80 \
     "${RAPIDS_TESTS_DIR}/junit-cuml-accel-scikit-learn.xml"
 

conda/environments/all_cuda-118_arch-aarch64.yaml

@@ -60,6 +60,7 @@ dependencies:
 - pytest-xdist
 - pytest==7.*
 - python>=3.10,<3.13
+- pyyaml
 - raft-dask==25.6.*,>=0.0.0a0
 - rapids-build-backend>=0.3.0,<0.4.0.dev0
 - rapids-dask-dependency==25.6.*,>=0.0.0a0

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -60,6 +60,7 @@ dependencies:
 - pytest-xdist
 - pytest==7.*
 - python>=3.10,<3.13
+- pyyaml
 - raft-dask==25.6.*,>=0.0.0a0
 - rapids-build-backend>=0.3.0,<0.4.0.dev0
 - rapids-dask-dependency==25.6.*,>=0.0.0a0

conda/environments/all_cuda-128_arch-aarch64.yaml

@@ -56,6 +56,7 @@ dependencies:
 - pytest-xdist
 - pytest==7.*
 - python>=3.10,<3.13
+- pyyaml
 - raft-dask==25.6.*,>=0.0.0a0
 - rapids-build-backend>=0.3.0,<0.4.0.dev0
 - rapids-dask-dependency==25.6.*,>=0.0.0a0

conda/environments/all_cuda-128_arch-x86_64.yaml

@@ -56,6 +56,7 @@ dependencies:
 - pytest-xdist
 - pytest==7.*
 - python>=3.10,<3.13
+- pyyaml
 - raft-dask==25.6.*,>=0.0.0a0
 - rapids-build-backend>=0.3.0,<0.4.0.dev0
 - rapids-dask-dependency==25.6.*,>=0.0.0a0

dependencies.yaml

@@ -486,6 +486,7 @@ dependencies:
           - hypothesis>=6.0,<7
           - nltk
           - numpydoc
+          - pyyaml
           - pytest==7.*
           - pytest-benchmark
           - pytest-cases

python/cuml/cuml/accel/__init__.py

@@ -17,12 +17,18 @@
 from cuml.accel.core import enabled, install
 from cuml.accel.estimator_proxy import is_proxy
 from cuml.accel.magics import load_ipython_extension
-from cuml.accel.pytest_plugin import pytest_load_initial_conftests
+from cuml.accel.pytest_plugin import (
+    pytest_addoption,
+    pytest_collection_modifyitems,
+    pytest_load_initial_conftests,
+)
 
 __all__ = (
     "enabled",
     "install",
     "is_proxy",
     "load_ipython_extension",
     "pytest_load_initial_conftests",
+    "pytest_collection_modifyitems",
+    "pytest_addoption",
 )

python/cuml/cuml/accel/pytest_plugin.py

@@ -14,6 +14,13 @@
 # limitations under the License.
 #
 
+from collections import defaultdict
+from importlib.metadata import version
+from pathlib import Path
+
+import yaml
+from packaging.requirements import Requirement
+
 from cuml.accel.core import install
 
 
@@ -26,3 +33,89 @@ def pytest_load_initial_conftests(early_config, parser, args):
         raise RuntimeError(
             "An existing plugin has already loaded sklearn. Interposing failed."
         )
+
+
+def pytest_addoption(parser):
+    """Add command line option for xfail list file."""
+    parser.addoption(
+        "--xfail-list",
+        action="store",
+        help="Path to YAML file containing list of test IDs to mark as xfail",
+    )
+
+
+def create_version_condition(condition_str: str) -> bool:
+    """Evaluate a version condition immediately.
+
+    Args:
+        condition_str: String in format 'package[comparison]version'
+                      For example:
+                      - 'scikit-learn>=1.5.2'
+                      - 'numpy<2.0.0'
+                      - 'pandas==2.1.0'
+
+    Returns:
+        bool: True if the condition is met, False otherwise
+    """
+    if not condition_str:
+        return True
+
+    try:
+        req = Requirement(condition_str)
+        installed_version = version(req.name)
+        return req.specifier.contains(installed_version)
+    except Exception:
+        return False
+
+
+def pytest_collection_modifyitems(config, items):
+    """Apply xfail markers to tests listed in the xfail list file."""
+    # Import pytest lazily to avoid requiring it for normal cuml usage.
+    # pytest is only needed when running tests.
+    import pytest
+
+    xfail_list_path = config.getoption("xfail_list")
+    if not xfail_list_path:
+        return
+
+    xfail_list_path = Path(xfail_list_path)
+    if not xfail_list_path.exists():
+        raise ValueError(f"Xfail list file not found: {xfail_list_path}")
+
+    xfail_list = yaml.safe_load(xfail_list_path.read_text())
+
+    if not isinstance(xfail_list, list):
+        raise ValueError("Xfail list must be a list of test entries")
+
+    # Convert list of dicts into dict mapping test IDs to lists of xfail configs
+    xfail_configs = defaultdict(list)
+    for entry in xfail_list:
+        if not isinstance(entry, dict):
+            raise ValueError("Xfail list entry must be a dictionary")
+        if "id" not in entry:
+            raise ValueError("Xfail list entry must contain an 'id' field")
+
+        test_id = entry["id"]
+        condition = True
+        if "condition" in entry:
+            condition = create_version_condition(entry["condition"])
+
+        config = {
+            "reason": entry.get("reason", "Test listed in xfail list"),
+            "strict": entry.get("strict", True),
+            "condition": condition,
+        }
+
+        xfail_configs[test_id].append(config)
+
+    for item in items:
+        test_id = f"{item.module.__name__}::{item.name}"
+        if test_id in xfail_configs:
+            for config in xfail_configs[test_id]:
+                item.add_marker(
+                    pytest.mark.xfail(
+                        reason=config["reason"],
+                        strict=config["strict"],
+                        condition=config["condition"],
+                    )
+                )