scikit-learn-intelex 2025.10.0__py313-none-win_amd64.whl

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/svm/_common.py

@@ -0,0 +1,403 @@
+# ==============================================================================
+# Copyright 2021 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 warnings
+from numbers import Number, Real
+
+import numpy as np
+from scipy import sparse as sp
+from sklearn.base import ClassifierMixin
+from sklearn.calibration import CalibratedClassifierCV
+from sklearn.metrics import r2_score
+from sklearn.preprocessing import LabelEncoder
+
+from daal4py.sklearn._utils import sklearn_check_version
+from daal4py.sklearn.utils.validation import get_requires_y_tag
+from onedal.utils.validation import _check_array, _column_or_1d
+
+from .._config import config_context, get_config
+from .._utils import PatchingConditionsChain
+from ..base import oneDALEstimator
+from ..utils.validation import validate_data
+
+if sklearn_check_version("1.6"):
+    from sklearn.calibration import _fit_calibrator
+    from sklearn.frozen import FrozenEstimator
+    from sklearn.utils import indexable
+    from sklearn.utils._response import _get_response_values
+    from sklearn.utils.multiclass import check_classification_targets
+    from sklearn.utils.validation import check_is_fitted
+
+    if sklearn_check_version("1.8"):
+        from ..utils._array_api import get_namespace
+
+    def _prefit_CalibratedClassifierCV_fit(self, X, y, **fit_params):
+        # This is a stop-gap solution where the cv='prefit' of CalibratedClassifierCV
+        # was removed and the single fold solution needs to be maintained. Discussion
+        # of the mathematical and performance implications of this choice can be found
+        # here: https://github.com/uxlfoundation/scikit-learn-intelex/pull/1879
+        # This is distilled from the sklearn CalibratedClassifierCV for sklearn <1.8 for
+        # use in sklearn > 1.8 to maintain performance.
+        check_classification_targets(y)
+        X, y = indexable(X, y)
+
+        estimator = self._get_estimator()
+
+        self.calibrated_classifiers_ = []
+        check_is_fitted(self.estimator, attributes=["classes_"])
+        self.classes_ = self.estimator.classes_
+
+        predictions, _ = _get_response_values(
+            estimator,
+            X,
+            response_method=["decision_function", "predict_proba"],
+        )
+        if predictions.ndim == 1:
+            # Reshape binary output from `(n_samples,)` to `(n_samples, 1)`
+            predictions = predictions.reshape(-1, 1)
+
+        if sklearn_check_version("1.8"):
+            xp, _ = get_namespace(X, y)
+            calibrated_classifier = _fit_calibrator(
+                estimator,
+                predictions,
+                y,
+                self.classes_,
+                self.method,
+                xp,
+            )
+        else:
+            calibrated_classifier = _fit_calibrator(
+                estimator,
+                predictions,
+                y,
+                self.classes_,
+                self.method,
+            )
+        self.calibrated_classifiers_.append(calibrated_classifier)
+
+        first_clf = self.calibrated_classifiers_[0].estimator
+        if hasattr(first_clf, "n_features_in_"):
+            self.n_features_in_ = first_clf.n_features_in_
+        if hasattr(first_clf, "feature_names_in_"):
+            self.feature_names_in_ = first_clf.feature_names_in_
+        return self
+
+
+class BaseSVM(oneDALEstimator):
+
+    @property
+    def _dual_coef_(self):
+        return self._dualcoef_
+
+    @_dual_coef_.setter
+    def _dual_coef_(self, value):
+        self._dualcoef_ = value
+        if hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator.dual_coef_ = value
+            if hasattr(self._onedal_estimator, "_onedal_model"):
+                del self._onedal_estimator._onedal_model
+
+    @_dual_coef_.deleter
+    def _dual_coef_(self):
+        del self._dualcoef_
+
+    @property
+    def intercept_(self):
+        return self._icept_
+
+    @intercept_.setter
+    def intercept_(self, value):
+        self._icept_ = value
+        if hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator.intercept_ = value
+            if hasattr(self._onedal_estimator, "_onedal_model"):
+                del self._onedal_estimator._onedal_model
+
+    @intercept_.deleter
+    def intercept_(self):
+        del self._icept_
+
+    def _onedal_gpu_supported(self, method_name, *data):
+        patching_status = PatchingConditionsChain(f"sklearn.{method_name}")
+        patching_status.and_conditions([(False, "GPU offloading is not supported.")])
+        return patching_status
+
+    def _onedal_cpu_supported(self, method_name, *data):
+        class_name = self.__class__.__name__
+        patching_status = PatchingConditionsChain(
+            f"sklearn.svm.{class_name}.{method_name}"
+        )
+        if method_name == "fit":
+            patching_status.and_conditions(
+                [
+                    (
+                        self.kernel in ["linear", "rbf", "poly", "sigmoid"],
+                        f'Kernel is "{self.kernel}" while '
+                        '"linear", "rbf", "poly" and "sigmoid" are only supported.',
+                    )
+                ]
+            )
+            return patching_status
+        inference_methods = (
+            ["predict", "score"]
+            if class_name.endswith("R")
+            else ["predict", "predict_proba", "decision_function", "score"]
+        )
+        if method_name in inference_methods:
+            patching_status.and_conditions(
+                [(hasattr(self, "_onedal_estimator"), "oneDAL model was not trained.")]
+            )
+            return patching_status
+        raise RuntimeError(f"Unknown method {method_name} in {class_name}")
+
+    def _compute_gamma_sigma(self, X):
+        # only run extended conversion if kernel is not linear
+        # set to a value = 1.0, so gamma will always be passed to
+        # the onedal estimator as a float type
+        if self.kernel == "linear":
+            return 1.0
+
+        if isinstance(self.gamma, str):
+            if self.gamma == "scale":
+                if sp.issparse(X):
+                    # var = E[X^2] - E[X]^2
+                    X_sc = (X.multiply(X)).mean() - (X.mean()) ** 2
+                else:
+                    X_sc = X.var()
+                _gamma = 1.0 / (X.shape[1] * X_sc) if X_sc != 0 else 1.0
+            elif self.gamma == "auto":
+                _gamma = 1.0 / X.shape[1]
+            else:
+                raise ValueError(
+                    "When 'gamma' is a string, it should be either 'scale' or "
+                    "'auto'. Got '{}' instead.".format(self.gamma)
+                )
+        else:
+            if sklearn_check_version("1.1") and not sklearn_check_version("1.2"):
+                if isinstance(self.gamma, Real):
+                    if self.gamma <= 0:
+                        msg = (
+                            f"gamma value must be > 0; {self.gamma!r} is invalid. Use"
+                            " a positive number or use 'auto' to set gamma to a"
+                            " value of 1 / n_features."
+                        )
+                        raise ValueError(msg)
+                    _gamma = self.gamma
+                else:
+                    msg = (
+                        "The gamma value should be set to 'scale', 'auto' or a"
+                        f" positive float value. {self.gamma!r} is not a valid option"
+                    )
+                    raise ValueError(msg)
+            else:
+                _gamma = self.gamma
+        return _gamma
+
+    def _onedal_fit_checks(self, X, y, sample_weight=None):
+        if hasattr(self, "decision_function_shape"):
+            if self.decision_function_shape not in ("ovr", "ovo", None):
+                raise ValueError(
+                    f"decision_function_shape must be either 'ovr' or 'ovo', "
+                    f"got {self.decision_function_shape}."
+                )
+
+        if y is None:
+            if get_requires_y_tag(self):
+                raise ValueError(
+                    f"This {self.__class__.__name__} estimator "
+                    f"requires y to be passed, but the target y is None."
+                )
+        # finite check occurs in onedal estimator
+        X, y = validate_data(
+            self,
+            X,
+            y,
+            dtype=[np.float64, np.float32],
+            ensure_all_finite=False,
+            accept_sparse="csr",
+        )
+        y = self._validate_targets(y)
+        sample_weight = self._get_sample_weight(X, y, sample_weight)
+        return X, y, sample_weight
+
+    def _get_sample_weight(self, X, y, sample_weight):
+        n_samples = X.shape[0]
+        dtype = X.dtype
+        if n_samples == 1:
+            raise ValueError("n_samples=1")
+
+        sample_weight = np.ascontiguousarray(
+            [] if sample_weight is None else sample_weight, dtype=np.float64
+        )
+
+        sample_weight_count = sample_weight.shape[0]
+        if sample_weight_count != 0 and sample_weight_count != n_samples:
+            raise ValueError(
+                "sample_weight and X have incompatible shapes: "
+                "%r vs %r\n"
+                "Note: Sparse matrices cannot be indexed w/"
+                "boolean masks (use `indices=True` in CV)."
+                % (len(sample_weight), X.shape)
+            )
+
+        if sample_weight_count == 0:
+            if not isinstance(self, ClassifierMixin) or self.class_weight_ is None:
+                return None
+            sample_weight = np.ones(n_samples, dtype=dtype)
+        elif isinstance(sample_weight, Number):
+            sample_weight = np.full(n_samples, sample_weight, dtype=dtype)
+        else:
+            sample_weight = _check_array(
+                sample_weight,
+                accept_sparse=False,
+                ensure_2d=False,
+                dtype=dtype,
+                order="C",
+            )
+            if sample_weight.ndim != 1:
+                raise ValueError("Sample weights must be 1D array or scalar")
+
+            if sample_weight.shape != (n_samples,):
+                raise ValueError(
+                    "sample_weight.shape == {}, expected {}!".format(
+                        sample_weight.shape, (n_samples,)
+                    )
+                )
+
+        if np.all(sample_weight <= 0):
+            if "nusvc" in self.__module__:
+                raise ValueError("negative dimensions are not allowed")
+            else:
+                raise ValueError(
+                    "Invalid input - all samples have zero or negative weights."
+                )
+
+        return sample_weight
+
+
+class BaseSVC(BaseSVM):
+    def _compute_balanced_class_weight(self, y):
+        y_ = _column_or_1d(y)
+        classes, _ = np.unique(y_, return_inverse=True)
+
+        le = LabelEncoder()
+        y_ind = le.fit_transform(y_)
+        if not np.isin(classes, le.classes_).all():
+            raise ValueError("classes should have valid labels that are in y")
+
+        recip_freq = len(y_) / (len(le.classes_) * np.bincount(y_ind).astype(np.float64))
+        return recip_freq[le.transform(classes)]
+
+    def _fit_proba(self, X, y, sample_weight=None, queue=None):
+        # TODO: rewrite this method when probabilities output is implemented in oneDAL
+
+        # LibSVM uses the random seed to control cross-validation for probability generation
+        # CalibratedClassifierCV with "prefit" does not use an RNG nor a seed. This may
+        # impact users without their knowledge, so display a warning.
+        if self.random_state is not None:
+            warnings.warn(
+                "random_state does not influence oneDAL SVM results",
+                RuntimeWarning,
+            )
+
+        params = self.get_params()
+        params["probability"] = False
+        params["decision_function_shape"] = "ovr"
+        clf_base = self.__class__(**params)
+
+        # We use stock metaestimators below, so the only way
+        # to pass a queue is using config_context.
+        cfg = get_config()
+        cfg["target_offload"] = queue
+        with config_context(**cfg):
+            clf_base.fit(X, y)
+
+            # Forced use of FrozenEstimator starting in sklearn 1.6
+            if sklearn_check_version("1.6"):
+                clf_base = FrozenEstimator(clf_base)
+
+                self.clf_prob = CalibratedClassifierCV(
+                    clf_base,
+                    ensemble=False,
+                    method="sigmoid",
+                )
+                # see custom stopgap solution defined above
+                _prefit_CalibratedClassifierCV_fit(self.clf_prob, X, y)
+            else:
+
+                self.clf_prob = CalibratedClassifierCV(
+                    clf_base,
+                    ensemble=False,
+                    cv="prefit",
+                    method="sigmoid",
+                ).fit(X, y)
+
+    def _save_attributes(self):
+        self.support_vectors_ = self._onedal_estimator.support_vectors_
+        self.n_features_in_ = self._onedal_estimator.n_features_in_
+        self.fit_status_ = 0
+        self.dual_coef_ = self._onedal_estimator.dual_coef_
+        self.shape_fit_ = self._onedal_estimator.class_weight_
+        self.classes_ = self._onedal_estimator.classes_
+        if isinstance(self, ClassifierMixin) or not sklearn_check_version("1.2"):
+            self.class_weight_ = self._onedal_estimator.class_weight_
+        self.support_ = self._onedal_estimator.support_
+
+        self._icept_ = self._onedal_estimator.intercept_
+        self._n_support = self._onedal_estimator._n_support
+        self._sparse = False
+        self._gamma = self._onedal_estimator._gamma
+        if self.probability:
+            length = int(len(self.classes_) * (len(self.classes_) - 1) / 2)
+            self._probA = np.zeros(length)
+            self._probB = np.zeros(length)
+        else:
+            self._probA = np.empty(0)
+            self._probB = np.empty(0)
+
+        self._dualcoef_ = self.dual_coef_
+
+        if sklearn_check_version("1.1"):
+            length = int(len(self.classes_) * (len(self.classes_) - 1) / 2)
+            self.n_iter_ = np.full((length,), self._onedal_estimator.n_iter_)
+
+
+class BaseSVR(BaseSVM):
+    def _save_attributes(self):
+        self.support_vectors_ = self._onedal_estimator.support_vectors_
+        self.n_features_in_ = self._onedal_estimator.n_features_in_
+        self.fit_status_ = 0
+        self.dual_coef_ = self._onedal_estimator.dual_coef_
+        self.shape_fit_ = self._onedal_estimator.shape_fit_
+        self.support_ = self._onedal_estimator.support_
+
+        self._icept_ = self._onedal_estimator.intercept_
+        self._n_support = [self.support_vectors_.shape[0]]
+        self._sparse = False
+        self._gamma = self._onedal_estimator._gamma
+        self._probA = None
+        self._probB = None
+
+        if sklearn_check_version("1.1"):
+            self.n_iter_ = self._onedal_estimator.n_iter_
+
+        self._dualcoef_ = self.dual_coef_
+
+    def _onedal_score(self, X, y, sample_weight=None, queue=None):
+        return r2_score(
+            y, self._onedal_predict(X, queue=queue), sample_weight=sample_weight
+        )

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/svm/nusvc.py

@@ -0,0 +1,278 @@
+# ==============================================================================
+# Copyright 2021 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.
+# ==============================================================================
+
+from functools import wraps
+
+import numpy as np
+from sklearn.exceptions import NotFittedError
+from sklearn.metrics import accuracy_score
+from sklearn.svm import NuSVC as _sklearn_NuSVC
+from sklearn.utils.metaestimators import available_if
+from sklearn.utils.validation import (
+    _deprecate_positional_args,
+    check_array,
+    check_is_fitted,
+)
+
+from daal4py.sklearn._n_jobs_support import control_n_jobs
+from daal4py.sklearn._utils import sklearn_check_version
+from onedal.svm import NuSVC as onedal_NuSVC
+
+from .._device_offload import dispatch, wrap_output_data
+from ..utils._array_api import get_namespace
+from ..utils.validation import validate_data
+from ._common import BaseSVC
+
+
+@control_n_jobs(
+    decorated_methods=["fit", "predict", "_predict_proba", "decision_function", "score"]
+)
+class NuSVC(BaseSVC, _sklearn_NuSVC):
+    __doc__ = _sklearn_NuSVC.__doc__
+
+    if sklearn_check_version("1.2"):
+        _parameter_constraints: dict = {**_sklearn_NuSVC._parameter_constraints}
+
+    @_deprecate_positional_args
+    def __init__(
+        self,
+        *,
+        nu=0.5,
+        kernel="rbf",
+        degree=3,
+        gamma="scale",
+        coef0=0.0,
+        shrinking=True,
+        probability=False,
+        tol=1e-3,
+        cache_size=200,
+        class_weight=None,
+        verbose=False,
+        max_iter=-1,
+        decision_function_shape="ovr",
+        break_ties=False,
+        random_state=None,
+    ):
+        super().__init__(
+            nu=nu,
+            kernel=kernel,
+            degree=degree,
+            gamma=gamma,
+            coef0=coef0,
+            shrinking=shrinking,
+            probability=probability,
+            tol=tol,
+            cache_size=cache_size,
+            class_weight=class_weight,
+            verbose=verbose,
+            max_iter=max_iter,
+            decision_function_shape=decision_function_shape,
+            break_ties=break_ties,
+            random_state=random_state,
+        )
+
+    def fit(self, X, y, sample_weight=None):
+        if sklearn_check_version("1.2"):
+            self._validate_params()
+        elif self.nu <= 0 or self.nu > 1:
+            # 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("nu <= 0 or nu > 1")
+        dispatch(
+            self,
+            "fit",
+            {
+                "onedal": self.__class__._onedal_fit,
+                "sklearn": _sklearn_NuSVC.fit,
+            },
+            X,
+            y,
+            sample_weight=sample_weight,
+        )
+
+        return self
+
+    @wrap_output_data
+    def predict(self, X):
+        check_is_fitted(self)
+        return dispatch(
+            self,
+            "predict",
+            {
+                "onedal": self.__class__._onedal_predict,
+                "sklearn": _sklearn_NuSVC.predict,
+            },
+            X,
+        )
+
+    @wrap_output_data
+    def score(self, X, y, sample_weight=None):
+        check_is_fitted(self)
+        return dispatch(
+            self,
+            "score",
+            {
+                "onedal": self.__class__._onedal_score,
+                "sklearn": _sklearn_NuSVC.score,
+            },
+            X,
+            y,
+            sample_weight=sample_weight,
+        )
+
+    @available_if(_sklearn_NuSVC._check_proba)
+    @wraps(_sklearn_NuSVC.predict_proba, assigned=["__doc__"])
+    def predict_proba(self, X):
+        check_is_fitted(self)
+        return self._predict_proba(X)
+
+    @available_if(_sklearn_NuSVC._check_proba)
+    @wraps(_sklearn_NuSVC.predict_log_proba, assigned=["__doc__"])
+    def predict_log_proba(self, X):
+        xp, _ = get_namespace(X)
+
+        return xp.log(self.predict_proba(X))
+
+    @wrap_output_data
+    def _predict_proba(self, X):
+        return dispatch(
+            self,
+            "predict_proba",
+            {
+                "onedal": self.__class__._onedal_predict_proba,
+                "sklearn": _sklearn_NuSVC.predict_proba,
+            },
+            X,
+        )
+
+    @wrap_output_data
+    def decision_function(self, X):
+        check_is_fitted(self)
+        return dispatch(
+            self,
+            "decision_function",
+            {
+                "onedal": self.__class__._onedal_decision_function,
+                "sklearn": _sklearn_NuSVC.decision_function,
+            },
+            X,
+        )
+
+    decision_function.__doc__ = _sklearn_NuSVC.decision_function.__doc__
+
+    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
+
+        weight_per_class = [
+            np.sum(sample_weight[y == class_label]) for class_label in np.unique(y)
+        ]
+
+        for i in range(len(weight_per_class)):
+            for j in range(i + 1, len(weight_per_class)):
+                if self.nu * (weight_per_class[i] + weight_per_class[j]) / 2 > min(
+                    weight_per_class[i], weight_per_class[j]
+                ):
+                    raise ValueError("specified nu is infeasible")
+
+        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 = {
+            "nu": self.nu,
+            "kernel": self.kernel,
+            "degree": self.degree,
+            "gamma": self._compute_gamma_sigma(X),
+            "coef0": self.coef0,
+            "tol": self.tol,
+            "shrinking": self.shrinking,
+            "cache_size": self.cache_size,
+            "max_iter": self.max_iter,
+            "class_weight": self.class_weight,
+            "break_ties": self.break_ties,
+            "decision_function_shape": self.decision_function_shape,
+        }
+
+        self._onedal_estimator = onedal_NuSVC(**onedal_params)
+        self._onedal_estimator.fit(X, y, weights, queue=queue)
+
+        if self.probability:
+            self._fit_proba(
+                X,
+                y,
+                sample_weight=sample_weight,
+                queue=queue,
+            )
+
+        self._save_attributes()
+
+    def _onedal_predict(self, X, queue=None):
+        validate_data(
+            self,
+            X,
+            dtype=[np.float64, np.float32],
+            ensure_all_finite=False,
+            ensure_2d=False,
+            accept_sparse="csr",
+            reset=False,
+        )
+
+        return self._onedal_estimator.predict(X, queue=queue)
+
+    def _onedal_predict_proba(self, X, queue=None):
+        if getattr(self, "clf_prob", None) is None:
+            raise NotFittedError(
+                "predict_proba is not available when fitted with probability=False"
+            )
+        from .._config import config_context, get_config
+
+        # We use stock metaestimators below, so the only way
+        # to pass a queue is using config_context.
+        cfg = get_config()
+        cfg["target_offload"] = queue
+        with config_context(**cfg):
+            return self.clf_prob.predict_proba(X)
+
+    def _onedal_decision_function(self, X, queue=None):
+        validate_data(
+            self,
+            X,
+            dtype=[np.float64, np.float32],
+            ensure_all_finite=False,
+            accept_sparse="csr",
+            reset=False,
+        )
+
+        return self._onedal_estimator.decision_function(X, queue=queue)
+
+    def _onedal_score(self, X, y, sample_weight=None, queue=None):
+        return accuracy_score(
+            y, self._onedal_predict(X, queue=queue), sample_weight=sample_weight
+        )
+
+    fit.__doc__ = _sklearn_NuSVC.fit.__doc__
+    predict.__doc__ = _sklearn_NuSVC.predict.__doc__
+    decision_function.__doc__ = _sklearn_NuSVC.decision_function.__doc__
+    score.__doc__ = _sklearn_NuSVC.score.__doc__