Remove deprecation warnings in 25.10

python/cuml/cuml/cluster/__init__.py

@@ -18,19 +18,3 @@
 from cuml.cluster.dbscan import DBSCAN
 from cuml.cluster.hdbscan import HDBSCAN
 from cuml.cluster.kmeans import KMeans
-
-
-def __getattr__(name):
-    import warnings
-
-    if name in ("all_points_membership_vectors", "approximate_predict"):
-        warnings.warn(
-            f"Accessing {name!r} from the `cuml.cluster` namespace is deprecated "
-            "and will be removed in 25.10. Please access it from the "
-            "`cuml.cluster.hdbscan` namespace instead.",
-            FutureWarning,
-        )
-        import cuml.cluster.hdbscan as mod
-
-        return getattr(mod, name)
-    raise AttributeError(f"module `cuml.cluster` has no attribute {name!r}")

python/cuml/cuml/cluster/hdbscan/__init__.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 #
 
-import cuml.cluster.hdbscan.prediction as prediction
 from cuml.cluster.hdbscan.hdbscan import (
     HDBSCAN,
     all_points_membership_vectors,

python/cuml/cuml/cluster/hdbscan/hdbscan.pyx

@@ -14,8 +14,6 @@
 #
 
 # distutils: language = c++
-import warnings
-
 import cupy as cp
 import numpy as np
 from pylibraft.common.handle import Handle
@@ -645,7 +643,6 @@ class HDBSCAN(Base, InteropMixin, ClusterMixin, CMajorInputTagMixin):
             "cluster_selection_method",
             "allow_single_cluster",
             "gen_min_span_tree",
-            "connectivity",
             "prediction_data"
         ]
 
@@ -754,7 +751,6 @@ class HDBSCAN(Base, InteropMixin, ClusterMixin, CMajorInputTagMixin):
                  gen_min_span_tree=False,
                  handle=None,
                  verbose=False,
-                 connectivity='deprecated',
                  output_type=None,
                  prediction_data=False):
 
@@ -765,12 +761,6 @@ class HDBSCAN(Base, InteropMixin, ClusterMixin, CMajorInputTagMixin):
         if min_samples is None:
             min_samples = min_cluster_size
 
-        if connectivity != "deprecated":
-            warnings.warn(
-                "The `connectivity` parameter is deprecated and will be removed in 25.10",
-                FutureWarning,
-            )
-
         if 2 < min_samples and min_samples > 1023:
             raise ValueError("'min_samples' must be a positive number "
                              "between 2 and 1023")
@@ -784,7 +774,6 @@ class HDBSCAN(Base, InteropMixin, ClusterMixin, CMajorInputTagMixin):
         self.alpha = alpha
         self.cluster_selection_method = cluster_selection_method
         self.allow_single_cluster = allow_single_cluster
-        self.connectivity = connectivity
         self.gen_min_span_tree = gen_min_span_tree
         self.prediction_data = prediction_data
 

python/cuml/cuml/fil/fil.pyx

@@ -15,7 +15,6 @@
 #
 import itertools
 import pathlib
-import warnings
 from time import perf_counter
 
 import numpy as np
@@ -621,7 +620,6 @@ class ForestInference(Base, CMajorInputTagMixin):
         output_type=None,
         verbose=False,
         is_classifier=False,
-        default_threshold=None,
         layout='depth_first',
         default_chunk_size=None,
         align_bytes=None,
@@ -631,9 +629,6 @@ class ForestInference(Base, CMajorInputTagMixin):
         super().__init__(
             handle=handle, verbose=verbose, output_type=output_type
         )
-        # TODO(hcho3): 25.10: Remove this parameter; direct users to set `threshold`
-        # in predict()
-        self.default_threshold = default_threshold if default_threshold else 0.5
         self.is_classifier = is_classifier
         self.default_chunk_size = default_chunk_size
         self.align_bytes = align_bytes
@@ -727,7 +722,6 @@ class ForestInference(Base, CMajorInputTagMixin):
         path,
         *,
         is_classifier=False,
-        threshold=None,
         precision='single',
         model_type=None,
         output_type=None,
@@ -749,8 +743,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             made to load the file based on its extension.
         is_classifier : boolean, default=False
             True for classification models, False for regressors
-        threshold : float
-            Deprecated in 25.08. Please set `threshold` in `predict()` instead.
         precision : {'single', 'double', None}, default='single'
             Use the given floating point precision for evaluating the model. If
             None, use the native precision of the model. Note that
@@ -791,12 +783,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             For GPU execution, the RAFT handle containing the stream or stream
             pool to use during loading and inference.
         """
-        if threshold is not None:
-            warnings.warn(
-                "The `threshold` parameter of `load()` is deprecated and "
-                "will be removed in 25.10. Please set `threshold` in `predict()` instead.",
-                FutureWarning
-            )
         if model_type is None:
             extension = pathlib.Path(path).suffix
             if extension == '.json':
@@ -827,7 +813,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             output_type=output_type,
             verbose=verbose,
             is_classifier=is_classifier,
-            default_threshold=threshold,
             default_chunk_size=default_chunk_size,
             align_bytes=align_bytes,
             layout=layout,
@@ -841,7 +826,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             skl_model,
             *,
             is_classifier=False,
-            threshold=None,
             precision='single',
             model_type=None,
             output_type=None,
@@ -859,8 +843,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             The Scikit-Learn forest model to load.
         is_classifier : boolean, default=False
             True for classification models, False for regressors
-        threshold : float
-            Deprecated in 25.08. Please set `threshold` in `predict()` instead.
         precision : {'single', 'double', None}, default='single'
             Use the given floating point precision for evaluating the model. If
             None, use the native precision of the model. Note that
@@ -911,20 +893,13 @@ class ForestInference(Base, CMajorInputTagMixin):
             For GPU execution, the RAFT handle containing the stream or stream
             pool to use during loading and inference.
         """
-        if threshold is not None:
-            warnings.warn(
-                "The `threshold` parameter of `load_from_sklearn()` is deprecated and "
-                "will be removed in 25.10. Please set `threshold` in `predict()` instead.",
-                FutureWarning
-            )
         tl_model = treelite.sklearn.import_model(skl_model)
         result = cls(
             treelite_model=tl_model,
             handle=handle,
             output_type=output_type,
             verbose=verbose,
             is_classifier=is_classifier,
-            default_threshold=threshold,
             default_chunk_size=default_chunk_size,
             align_bytes=align_bytes,
             layout=layout,
@@ -939,7 +914,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             tl_model,
             *,
             is_classifier=False,
-            threshold=None,
             precision='single',
             model_type=None,
             output_type=None,
@@ -957,8 +931,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             The Treelite model to load.
         is_classifier : boolean, default=False
             True for classification models, False for regressors
-        threshold : float
-            Deprecated in 25.08. Please set `threshold` in `predict()` instead.
         precision : {'single', 'double', None}, default='single'
             Use the given floating point precision for evaluating the model. If
             None, use the native precision of the model. Note that
@@ -1009,19 +981,12 @@ class ForestInference(Base, CMajorInputTagMixin):
             For GPU execution, the RAFT handle containing the stream or stream
             pool to use during loading and inference.
         """
-        if threshold is not None:
-            warnings.warn(
-                "The `threshold` parameter of `load_from_treelite_model()` is deprecated and "
-                "will be removed in 25.10. Please set `threshold` in `predict()` instead.",
-                FutureWarning
-            )
         return cls(
             treelite_model=tl_model,
             handle=handle,
             output_type=output_type,
             verbose=verbose,
             is_classifier=is_classifier,
-            default_threshold=threshold,
             default_chunk_size=default_chunk_size,
             align_bytes=align_bytes,
             layout=layout,
@@ -1156,7 +1121,7 @@ class ForestInference(Base, CMajorInputTagMixin):
             proba = self.forest.predict(X, chunk_size=chunk_size)
             if len(proba.shape) < 2 or proba.shape[1] == 1:
                 if threshold is None:
-                    threshold = self.default_threshold
+                    threshold = 0.5
                 result = (
                     proba.to_output(output_type='array') > threshold
                 ).astype('int')
@@ -1413,7 +1378,6 @@ class ForestInference(Base, CMajorInputTagMixin):
             "output_type",
             "verbose",
             "is_classifier",
-            "default_threshold",
             "layout",
             "default_chunk_size",
             "align_bytes",

python/cuml/cuml/solvers/sgd.pyx

@@ -14,8 +14,6 @@
 #
 
 # distutils: language = c++
-import warnings
-
 import cupy as cp
 import numpy as np
 
@@ -230,14 +228,7 @@ class SGD(Base,
         else:
             raise ValueError(f"loss {loss!r} is not supported")
 
-        if penalty == 'none':
-            warnings.warn(
-                "penalty='none' is deprecated and will be removed in 25.10. Please use "
-                "`penalty=None` instead.",
-                FutureWarning,
-            )
-            self.penalty = None
-        elif penalty in [None, 'l1', 'l2', 'elasticnet']:
+        if penalty in [None, 'l1', 'l2', 'elasticnet']:
             self.penalty = penalty
         else:
             raise ValueError(f"penalty {penalty!r} is not supported")

python/cuml/tests/test_fil.py

@@ -552,29 +552,6 @@ def test_chunk_size(chunk_size, small_classifier_and_preds):
     np.testing.assert_array_equal(fil_preds, xgb_preds_int)
 
 
-@pytest.mark.parametrize("is_classifier", [True, False])
-def test_thresholding(is_classifier, small_classifier_and_preds):
-    model_path, model_type, X, xgb_preds = small_classifier_and_preds
-    fm = ForestInference.load(
-        model_path,
-        model_type=model_type,
-        is_classifier=is_classifier,
-    )
-    fil_preds = np.asarray(fm.predict(X, threshold=0.5))
-    if is_classifier:
-        assert ((fil_preds != 0.0) & (fil_preds != 1.0)).sum() == 0
-    else:
-        assert ((fil_preds != 0.0) & (fil_preds != 1.0)).sum() > 0
-
-    with pytest.warns(FutureWarning):
-        _ = ForestInference.load(
-            model_path,
-            model_type=model_type,
-            is_classifier=is_classifier,
-            threshold=0.5,
-        )
-
-
 @pytest.mark.parametrize("train_device", ("cpu", "gpu"))
 @pytest.mark.parametrize("infer_device", ("cpu", "gpu"))
 def test_output_args(train_device, infer_device, small_classifier_and_preds):

python/cuml/tests/test_hdbscan.py

@@ -1144,44 +1144,6 @@ def test_membership_vector_circles(
     assert_membership_vectors(cu_membership_vectors, sk_membership_vectors)
 
 
-def test_prediction_namespace_deprecated():
-    # Attribute access warns
-    with pytest.warns(FutureWarning, match="all_points_membership_vectors"):
-        func = cuml.cluster.hdbscan.prediction.all_points_membership_vectors
-
-    assert func is cuml.cluster.hdbscan.all_points_membership_vectors
-
-    # Imports warn
-    with pytest.warns(FutureWarning, match="all_points_membership_vectors"):
-        from cuml.cluster.hdbscan.prediction import (
-            all_points_membership_vectors as func,
-        )
-
-    assert func is cuml.cluster.hdbscan.all_points_membership_vectors
-
-    # Unknown attribute errors
-    with pytest.raises(AttributeError, match="not_a_real_attr"):
-        cuml.cluster.hdbscan.prediction.not_a_real_attr
-
-
-def test_prediction_functions_cluster_namespace_deprecated():
-    # Attribute access warns
-    with pytest.warns(FutureWarning, match="all_points_membership_vectors"):
-        func = cuml.cluster.all_points_membership_vectors
-
-    assert func is cuml.cluster.hdbscan.all_points_membership_vectors
-
-    # Imports warn
-    with pytest.warns(FutureWarning, match="all_points_membership_vectors"):
-        from cuml.cluster import all_points_membership_vectors as func
-
-    assert func is cuml.cluster.hdbscan.all_points_membership_vectors
-
-    # Unknown attribute errors
-    with pytest.raises(AttributeError, match="not_a_real_attr"):
-        cuml.cluster.not_a_real_attr
-
-
 def test_all_points_membership_vectors_output_type():
     X, y = make_blobs(random_state=42)
     X_df = pd.DataFrame(X, index=[f"row{i}" for i in range(X.shape[0])])

python/cuml/tests/test_sgd.py

@@ -22,12 +22,6 @@
 from cuml.solvers import SGD as cumlSGD
 
 
-def test_sgd_penalty_none_deprecated():
-    with pytest.warns(FutureWarning, match="penalty='none' is deprecated"):
-        solver = cumlSGD(penalty="none")
-    assert solver.penalty is None
-
-
 @pytest.mark.parametrize("lrate", ["constant", "invscaling", "adaptive"])
 @pytest.mark.parametrize("dtype", [np.float32, np.float64])
 @pytest.mark.parametrize("penalty", [None, "l1", "l2", "elasticnet"])