Fix scikit-learn 1.8 test incompatibilities

conda/environments/all_cuda-129_arch-aarch64.yaml

@@ -71,7 +71,6 @@ dependencies:
 - rmm==26.2.*,>=0.0.0a0
 - scikit-build-core>=0.10.0
 - scikit-learn>=1.4
-- scikit-learn>=1.4,<1.8.0
 - scipy>=1.11.0
 - seaborn
 - sphinx

conda/environments/all_cuda-129_arch-x86_64.yaml

@@ -71,7 +71,6 @@ dependencies:
 - rmm==26.2.*,>=0.0.0a0
 - scikit-build-core>=0.10.0
 - scikit-learn>=1.4
-- scikit-learn>=1.4,<1.8.0
 - scipy>=1.11.0
 - seaborn
 - sphinx

conda/environments/all_cuda-130_arch-aarch64.yaml

@@ -71,7 +71,6 @@ dependencies:
 - rmm==26.2.*,>=0.0.0a0
 - scikit-build-core>=0.10.0
 - scikit-learn>=1.4
-- scikit-learn>=1.4,<1.8.0
 - scipy>=1.11.0
 - seaborn
 - sphinx

conda/environments/all_cuda-130_arch-x86_64.yaml

@@ -71,7 +71,6 @@ dependencies:
 - rmm==26.2.*,>=0.0.0a0
 - scikit-build-core>=0.10.0
 - scikit-learn>=1.4
-- scikit-learn>=1.4,<1.8.0
 - scipy>=1.11.0
 - seaborn
 - sphinx

dependencies.yaml

@@ -486,7 +486,7 @@ dependencies:
           - pytest-cov
           - pytest-xdist
           - seaborn
-          - scikit-learn>=1.4,<1.8.0
+          - *scikit_learn
           - statsmodels
           - tenacity
           - umap-learn==0.5.7

python/cuml/cuml_accel_tests/integration/test_hdbscan.py

@@ -13,6 +13,9 @@
 from sklearn.preprocessing import StandardScaler
 
 if Version(sklearn.__version__) >= Version("1.8.0.dev0"):
+    # NOTE: Remove this skip when issue
+    # https://github.com/scikit-learn-contrib/hdbscan/issues/689 is resolved,
+    # as it blocks compatibility with scikit-learn >= 1.8.0.dev0.
     pytest.skip(
         "hdbscan requires sklearn < 1.8.0.dev0", allow_module_level=True
     )

python/cuml/cuml_accel_tests/upstream/umap/run-tests.sh

@@ -14,6 +14,21 @@ set -eu
 
 UMAP_TAG="release-0.5.7"
 
+# Skip tests for scikit-learn >= 1.8 -- umap-learn is not compatible with scikit-learn 1.8 yet
+python -c "
+import sys
+from packaging.version import Version
+import sklearn
+sys.exit(
+    int(
+        Version(sklearn.__version__) >= Version('1.8')
+    )
+)
+" || {
+    echo "Skipping umap tests for scikit-learn >= 1.8"
+    exit 0
+}
+
 THIS_DIRECTORY=$( cd "$(dirname "${BASH_SOURCE[0]}")" ; pwd -P )
 UMAP_REPO="${THIS_DIRECTORY}/umap-upstream"
 

python/cuml/pyproject.toml

@@ -120,7 +120,7 @@ test = [
     "pytest-xdist",
     "pytest<9.0",
     "pyyaml",
-    "scikit-learn>=1.4,<1.8.0",
+    "scikit-learn>=1.4",
     "seaborn",
     "statsmodels",
     "tenacity",

python/cuml/tests/test_fil.py

@@ -10,7 +10,9 @@
 import numpy as np
 import pandas as pd
 import pytest
+import sklearn
 import treelite
+from packaging.version import Version
 
 # Import XGBoost before scikit-learn to work around a libgomp bug
 # See https://github.com/dmlc/xgboost/issues/7110
@@ -910,6 +912,9 @@ def test_device_selection(device_id, model_kind, tmp_path):
         )
         xgb_model.fit(X, y)
         model_path = os.path.join(tmp_path, "xgb_class.ubj")
+        # skip with sklearn version 1.8.0.dev0
+        if Version(sklearn.__version__) >= Version("1.8.0.dev0"):
+            pytest.skip("xgboost is incompatible with sklearn >= 1.8.0.dev0")
         xgb_model.save_model(model_path)
         fm = ForestInference.load(
             model_path,

python/cuml/tests/test_hdbscan.py

@@ -6,6 +6,8 @@
 import numpy as np
 import pandas as pd
 import pytest
+import sklearn
+from packaging.version import Version
 from pylibraft.common import DeviceResourcesSNMG
 from sklearn import datasets
 from sklearn.datasets import make_blobs
@@ -23,6 +25,11 @@
 from cuml.testing.datasets import make_pattern
 from cuml.testing.utils import array_equal
 
+if Version(sklearn.__version__) >= Version("1.8.0.dev0"):
+    pytest.skip(
+        "hdbscan requires sklearn < 1.8.0.dev0", allow_module_level=True
+    )
+
 dataset_names = ["noisy_circles", "noisy_moons", "varied"]
 
 

python/cuml/tests/test_linear_model.py

@@ -329,23 +329,20 @@ def test_logistic_regression(
                 solver="saga",
                 C=C,
                 fit_intercept=fit_intercept,
-                multi_class="auto",
             )
         else:
             sklog = skLog(
                 penalty=penalty,
                 solver="saga",
                 C=C,
                 fit_intercept=fit_intercept,
-                multi_class="auto",
             )
     else:
         sklog = skLog(
             penalty=penalty,
             solver="lbfgs",
             C=C,
             fit_intercept=fit_intercept,
-            multi_class="auto",
         )
 
     sklog.fit(X_train, y_train)
@@ -417,7 +414,7 @@ def test_logistic_regression_model_default(dtype):
     y_test = y_test.astype(dtype)
     culog = cuLog()
     culog.fit(X_train, y_train)
-    sklog = skLog(multi_class="auto")
+    sklog = skLog()
 
     sklog.fit(X_train, y_train)
 
@@ -586,11 +583,7 @@ def test_logistic_regression_predict_proba(
 
     sklog = skLog(
         fit_intercept=fit_intercept,
-        **(
-            {"solver": "lbfgs", "multi_class": "multinomial"}
-            if num_classes > 2
-            else {}
-        ),
+        **({"solver": "lbfgs"} if num_classes > 2 else {}),
     )
     sklog.coef_ = culog.coef_
     sklog.intercept_ = culog.intercept_ if fit_intercept else 0

python/cuml/tests/test_trustworthiness.py

@@ -4,6 +4,8 @@
 import cudf
 import numpy as np
 import pytest
+import sklearn
+from packaging.version import Version
 from sklearn.datasets import make_blobs
 from sklearn.manifold import trustworthiness as sklearn_trustworthiness
 from umap import UMAP
@@ -19,6 +21,10 @@
 @pytest.mark.filterwarnings(
     "ignore:n_jobs value.*overridden.*by setting random_state.*:UserWarning"
 )
+@pytest.mark.xfail(
+    condition=Version(sklearn.__version__) >= Version("1.8.0.dev0"),
+    reason="umap-learn is incompatible with sklearn >= 1.8.0",
+)
 def test_trustworthiness(
     input_type, n_samples, n_features, n_components, batch_size
 ):

python/cuml/tests/test_umap.py

@@ -9,7 +9,9 @@
 import numpy as np
 import pytest
 import scipy.sparse as scipy_sparse
+import sklearn
 import umap
+from packaging.version import Version
 from pylibraft.common import DeviceResourcesSNMG
 from sklearn import datasets
 from sklearn.cluster import KMeans
@@ -29,6 +31,9 @@
     unit_param,
 )
 
+if Version(sklearn.__version__) >= Version("1.8.0.dev0"):
+    pytest.skip("umap requires sklearn < 1.8.0.dev0", allow_module_level=True)
+
 dataset_names = ["iris", "digits", "wine", "blobs"]