scikit-learn-intelex 2024.4.0__py311-none-manylinux1_x86_64.whl → 2024.6.0__py311-none-manylinux1_x86_64.whl

sklearnex/svm/svc.py

@@ -85,6 +85,17 @@ class SVC(sklearn_SVC, BaseSVC):
     def fit(self, X, y, sample_weight=None):
         if sklearn_check_version("1.2"):
             self._validate_params()
+        elif self.C <= 0:
+            # else if added to correct issues with
+            # sklearn tests:
+            # svm/tests/test_sparse.py::test_error
+            # svm/tests/test_svm.py::test_bad_input
+            # for sklearn versions < 1.2 (i.e. without
+            # validate_params parameter checking)
+            # Without this, a segmentation fault with
+            # Windows fatal exception: access violation
+            # occurs
+            raise ValueError("C <= 0")
         if sklearn_check_version("1.0"):
             self._check_feature_names(X, reset=True)
         dispatch(
@@ -96,8 +107,9 @@ class SVC(sklearn_SVC, BaseSVC):
             },
             X,
             y,
-            sample_weight,
+            sample_weight=sample_weight,
         )
+
         return self
 
     @wrap_output_data
@@ -270,12 +282,30 @@ class SVC(sklearn_SVC, BaseSVC):
             return patching_status
         raise RuntimeError(f"Unknown method {method_name} in {class_name}")
 
+    def _get_sample_weight(self, X, y, sample_weight=None):
+        sample_weight = super()._get_sample_weight(X, y, sample_weight)
+        if sample_weight is None:
+            return sample_weight
+
+        if np.any(sample_weight <= 0) and len(np.unique(y[sample_weight > 0])) != len(
+            self.classes_
+        ):
+            raise ValueError(
+                "Invalid input - all samples with positive weights "
+                "belong to the same class"
+                if sklearn_check_version("1.2")
+                else "Invalid input - all samples with positive weights "
+                "have the same label."
+            )
+        return sample_weight
+
     def _onedal_fit(self, X, y, sample_weight=None, queue=None):
+        X, _, weights = self._onedal_fit_checks(X, y, sample_weight)
         onedal_params = {
             "C": self.C,
             "kernel": self.kernel,
             "degree": self.degree,
-            "gamma": self.gamma,
+            "gamma": self._compute_gamma_sigma(X),
             "coef0": self.coef0,
             "tol": self.tol,
             "shrinking": self.shrinking,
@@ -287,10 +317,16 @@ class SVC(sklearn_SVC, BaseSVC):
         }
 
         self._onedal_estimator = onedal_SVC(**onedal_params)
-        self._onedal_estimator.fit(X, y, sample_weight, queue=queue)
+        self._onedal_estimator.fit(X, y, weights, queue=queue)
 
         if self.probability:
-            self._fit_proba(X, y, sample_weight, queue=queue)
+            self._fit_proba(
+                X,
+                y,
+                sample_weight=sample_weight,
+                queue=queue,
+            )
+
         self._save_attributes()
 
     def _onedal_predict(self, X, queue=None):

sklearnex/svm/svr.py

@@ -65,6 +65,17 @@ class SVR(sklearn_SVR, BaseSVR):
     def fit(self, X, y, sample_weight=None):
         if sklearn_check_version("1.2"):
             self._validate_params()
+        elif self.C <= 0:
+            # else if added to correct issues with
+            # sklearn tests:
+            # svm/tests/test_sparse.py::test_error
+            # svm/tests/test_svm.py::test_bad_input
+            # for sklearn versions < 1.2 (i.e. without
+            # validate_params parameter checking)
+            # Without this, a segmentation fault with
+            # Windows fatal exception: access violation
+            # occurs
+            raise ValueError("C <= 0")
         if sklearn_check_version("1.0"):
             self._check_feature_names(X, reset=True)
         dispatch(
@@ -76,7 +87,7 @@ class SVR(sklearn_SVR, BaseSVR):
             },
             X,
             y,
-            sample_weight,
+            sample_weight=sample_weight,
         )
 
         return self
@@ -95,13 +106,30 @@ class SVR(sklearn_SVR, BaseSVR):
             X,
         )
 
+    @wrap_output_data
+    def score(self, X, y, sample_weight=None):
+        if sklearn_check_version("1.0"):
+            self._check_feature_names(X, reset=False)
+        return dispatch(
+            self,
+            "score",
+            {
+                "onedal": self.__class__._onedal_score,
+                "sklearn": sklearn_SVR.score,
+            },
+            X,
+            y,
+            sample_weight=sample_weight,
+        )
+
     def _onedal_fit(self, X, y, sample_weight=None, queue=None):
+        X, _, sample_weight = self._onedal_fit_checks(X, y, sample_weight)
         onedal_params = {
             "C": self.C,
             "epsilon": self.epsilon,
             "kernel": self.kernel,
             "degree": self.degree,
-            "gamma": self.gamma,
+            "gamma": self._compute_gamma_sigma(X),
             "coef0": self.coef0,
             "tol": self.tol,
             "shrinking": self.shrinking,
@@ -118,3 +146,4 @@ class SVR(sklearn_SVR, BaseSVR):
 
     fit.__doc__ = sklearn_SVR.fit.__doc__
     predict.__doc__ = sklearn_SVR.predict.__doc__
+    score.__doc__ = sklearn_SVR.score.__doc__

sklearnex/tests/_utils.py

@@ -14,9 +14,12 @@
 # limitations under the License.
 # ==============================================================================
 
+from functools import partial
 from inspect import isclass
 
 import numpy as np
+from scipy import sparse as sp
+from sklearn import clone
 from sklearn.base import (
     BaseEstimator,
     ClassifierMixin,
@@ -87,18 +90,26 @@ mixin_map = [
 ]
 
 
-SPECIAL_INSTANCES = {
-    str(i): i
-    for i in [
-        LocalOutlierFactor(novelty=True),
-        SVC(probability=True),
-        NuSVC(probability=True),
-        KNeighborsClassifier(algorithm="brute"),
-        KNeighborsRegressor(algorithm="brute"),
-        NearestNeighbors(algorithm="brute"),
-        LogisticRegression(solver="newton-cg"),
-    ]
-}
+class _sklearn_clone_dict(dict):
+
+    def __getitem__(self, key):
+        return clone(super().__getitem__(key))
+
+
+SPECIAL_INSTANCES = _sklearn_clone_dict(
+    {
+        str(i): i
+        for i in [
+            LocalOutlierFactor(novelty=True),
+            SVC(probability=True),
+            NuSVC(probability=True),
+            KNeighborsClassifier(algorithm="brute"),
+            KNeighborsRegressor(algorithm="brute"),
+            NearestNeighbors(algorithm="brute"),
+            LogisticRegression(solver="newton-cg"),
+        ]
+    }
+)
 
 
 def gen_models_info(algorithms):
@@ -107,8 +118,8 @@ def gen_models_info(algorithms):
 
         if i in PATCHED_MODELS:
             est = PATCHED_MODELS[i]
-        elif i in SPECIAL_INSTANCES:
-            est = SPECIAL_INSTANCES[i].__class__
+        elif isinstance(algorithms[i], BaseEstimator):
+            est = algorithms[i].__class__
         else:
             raise KeyError(f"Unrecognized sklearnex estimator: {i}")
 
@@ -129,24 +140,54 @@ def gen_models_info(algorithms):
     return output
 
 
-def gen_dataset(estimator, queue=None, target_df=None, dtype=np.float64):
-    dataset = None
-    name = estimator.__class__.__name__
-    est = PATCHED_MODELS[name]
+def gen_dataset_type(est):
+    # est should be an estimator or estimator class
+    # dataset initialized to classification, but will be swapped
+    # for other types as necessary
+    dataset = "classification"
+    estimator = est.__class__ if isinstance(est, BaseEstimator) else est
+
     for mixin, _, data in mixin_map:
-        if issubclass(est, mixin) and data is not None:
+        if issubclass(estimator, mixin) and data is not None:
             dataset = data
+    return dataset
+
+
+_dataset_dict = {
+    "classification": [partial(load_iris, return_X_y=True)],
+    "regression": [partial(load_diabetes, return_X_y=True)],
+}
+
+
+def gen_dataset(
+    est,
+    datasets=_dataset_dict,
+    sparse=False,
+    queue=None,
+    target_df=None,
+    dtype=None,
+):
+    dataset_type = gen_dataset_type(est)
+    output = []
     # load data
-    if dataset == "classification" or dataset is None:
-        X, y = load_iris(return_X_y=True)
-    elif dataset == "regression":
-        X, y = load_diabetes(return_X_y=True)
-    else:
-        raise ValueError("Unknown dataset type")
-
-    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=target_df, dtype=dtype)
-    y = _convert_to_dataframe(y, sycl_queue=queue, target_df=target_df, dtype=dtype)
-    return X, y
+    flag = dtype is None
+
+    for func in datasets[dataset_type]:
+        X, y = func()
+        if flag:
+            dtype = X.dtype if hasattr(X, "dtype") else np.float64
+
+        if sparse:
+            X = sp.csr_matrix(X)
+        else:
+            X = _convert_to_dataframe(
+                X, sycl_queue=queue, target_df=target_df, dtype=dtype
+            )
+            y = _convert_to_dataframe(
+                y, sycl_queue=queue, target_df=target_df, dtype=dtype
+            )
+        output += [[X, y]]
+    return output
 
 
 DTYPES = [

sklearnex/tests/test_common.py

@@ -0,0 +1,54 @@
+# ==============================================================================
+# Copyright 2024 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+import os
+from glob import glob
+
+import pytest
+
+ALLOWED_LOCATIONS = [
+    "_config.py",
+    "_device_offload.py",
+    "test",
+    "svc.py",
+    "svm" + os.sep + "_common.py",
+]
+
+
+def test_target_offload_ban():
+    """This test blocks the use of target_offload in
+    in sklearnex files. Offloading computation to devices
+    via target_offload should only occur externally, and not
+    within the architecture of the sklearnex classes. This
+    is for clarity, traceability and maintainability.
+    """
+    from sklearnex import __file__ as loc
+
+    path = loc.replace("__init__.py", "")
+    files = [y for x in os.walk(path) for y in glob(os.path.join(x[0], "*.py"))]
+
+    output = []
+
+    for f in files:
+        if open(f, "r").read().find("target_offload") != -1:
+            output += [f.replace(path, "sklearnex" + os.sep)]
+
+    # remove this file from the list
+    for allowed in ALLOWED_LOCATIONS:
+        output = [i for i in output if allowed not in i]
+
+    output = "\n".join(output)
+    assert output == "", f"sklearn versioning is occuring in: \n{output}"