scikit-learn-intelex 2025.0.0__py312-none-manylinux_2_28_x86_64.whl

sklearnex/decomposition/pca.py

@@ -0,0 +1,414 @@
+# ===============================================================================
+# 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 logging
+
+from daal4py.sklearn._utils import daal_check_version
+
+if daal_check_version((2024, "P", 100)):
+    import numbers
+    from math import sqrt
+    from warnings import warn
+
+    import numpy as np
+    from scipy.sparse import issparse
+    from sklearn.utils.validation import check_is_fitted
+
+    from daal4py.sklearn._n_jobs_support import control_n_jobs
+    from daal4py.sklearn._utils import sklearn_check_version
+
+    from .._device_offload import dispatch, wrap_output_data
+    from .._utils import PatchingConditionsChain
+    from ..utils._array_api import get_namespace
+
+    if sklearn_check_version("1.1") and not sklearn_check_version("1.2"):
+        from sklearn.utils import check_scalar
+
+    if sklearn_check_version("1.2"):
+        from sklearn.utils._param_validation import StrOptions
+
+    from sklearn.decomposition import PCA as sklearn_PCA
+
+    from onedal.decomposition import PCA as onedal_PCA
+
+    @control_n_jobs(decorated_methods=["fit", "transform", "fit_transform"])
+    class PCA(sklearn_PCA):
+        __doc__ = sklearn_PCA.__doc__
+
+        if sklearn_check_version("1.2"):
+            _parameter_constraints: dict = {**sklearn_PCA._parameter_constraints}
+            # "onedal_svd" solver uses oneDAL's PCA-SVD algorithm
+            # and required for testing purposes to fully enable it in future.
+            # "covariance_eigh" solver is added for ability to explicitly request
+            # oneDAL's PCA-Covariance algorithm using any sklearn version < 1.5.
+            _parameter_constraints["svd_solver"] = [
+                StrOptions(
+                    _parameter_constraints["svd_solver"][0].options
+                    | {"onedal_svd", "covariance_eigh"}
+                )
+            ]
+
+        if sklearn_check_version("1.1"):
+
+            def __init__(
+                self,
+                n_components=None,
+                *,
+                copy=True,
+                whiten=False,
+                svd_solver="auto",
+                tol=0.0,
+                iterated_power="auto",
+                n_oversamples=10,
+                power_iteration_normalizer="auto",
+                random_state=None,
+            ):
+                self.n_components = n_components
+                self.copy = copy
+                self.whiten = whiten
+                self.svd_solver = svd_solver
+                self.tol = tol
+                self.iterated_power = iterated_power
+                self.n_oversamples = n_oversamples
+                self.power_iteration_normalizer = power_iteration_normalizer
+                self.random_state = random_state
+
+        else:
+
+            def __init__(
+                self,
+                n_components=None,
+                copy=True,
+                whiten=False,
+                svd_solver="auto",
+                tol=0.0,
+                iterated_power="auto",
+                random_state=None,
+            ):
+                self.n_components = n_components
+                self.copy = copy
+                self.whiten = whiten
+                self.svd_solver = svd_solver
+                self.tol = tol
+                self.iterated_power = iterated_power
+                self.random_state = random_state
+
+        def fit(self, X, y=None):
+            self._fit(X)
+            return self
+
+        @wrap_output_data
+        def _fit(self, X):
+            if sklearn_check_version("1.2"):
+                self._validate_params()
+            elif sklearn_check_version("1.1"):
+                check_scalar(
+                    self.n_oversamples,
+                    "n_oversamples",
+                    min_val=1,
+                    target_type=numbers.Integral,
+                )
+
+            return dispatch(
+                self,
+                "fit",
+                {
+                    "onedal": self.__class__._onedal_fit,
+                    "sklearn": sklearn_PCA._fit,
+                },
+                X,
+            )
+
+        def _onedal_fit(self, X, queue=None):
+            X = self._validate_data(
+                X,
+                dtype=[np.float64, np.float32],
+                ensure_2d=True,
+                copy=self.copy,
+            )
+
+            onedal_params = {
+                "n_components": self.n_components,
+                "is_deterministic": True,
+                "method": "svd" if self._fit_svd_solver == "onedal_svd" else "cov",
+                "whiten": self.whiten,
+            }
+            self._onedal_estimator = onedal_PCA(**onedal_params)
+            self._onedal_estimator.fit(X, queue=queue)
+            self._save_attributes()
+
+            U = None
+            S = self.singular_values_
+            Vt = self.components_
+
+            if sklearn_check_version("1.5"):
+                xp, _ = get_namespace(X)
+                x_is_centered = not self.copy
+
+                return U, S, Vt, X, x_is_centered, xp
+            else:
+                return U, S, Vt
+
+        @wrap_output_data
+        def transform(self, X):
+            return dispatch(
+                self,
+                "transform",
+                {
+                    "onedal": self.__class__._onedal_transform,
+                    "sklearn": sklearn_PCA.transform,
+                },
+                X,
+            )
+
+        def _onedal_transform(self, X, queue=None):
+            check_is_fitted(self)
+            if sklearn_check_version("1.0"):
+                self._check_feature_names(X, reset=False)
+            X = self._validate_data(
+                X,
+                dtype=[np.float64, np.float32],
+                reset=False,
+            )
+            self._validate_n_features_in_after_fitting(X)
+
+            return self._onedal_estimator.predict(X, queue=queue)
+
+        def fit_transform(self, X, y=None):
+            if sklearn_check_version("1.5"):
+                U, S, Vt, X_fit, x_is_centered, xp = self._fit(X)
+            else:
+                U, S, Vt = self._fit(X)
+                X_fit = X
+            if hasattr(self, "_onedal_estimator"):
+                # oneDAL PCA was fit
+                return self.transform(X)
+            elif U is not None:
+                # Scikit-learn PCA was fit
+                U = U[:, : self.n_components_]
+
+                if self.whiten:
+                    U *= sqrt(X_fit.shape[0] - 1)
+                else:
+                    U *= S[: self.n_components_]
+
+                return U
+            else:
+                # Scikit-learn PCA["covariance_eigh"] was fit
+                return self._transform(X_fit, xp, x_is_centered=x_is_centered)
+
+        @wrap_output_data
+        def inverse_transform(self, X):
+            xp, _ = get_namespace(X)
+
+            mean = self.mean_
+            if self.whiten:
+                components = (
+                    xp.sqrt(self.explained_variance_[:, np.newaxis]) * self.components_
+                )
+            else:
+                components = self.components_
+
+            if "numpy" not in xp.__name__:
+                # DPCtl and dpnp require inputs to be on the same device for
+                # matrix multiplication and division. The type and location
+                # of the components and mean are dependent on the sklearn
+                # version, this makes sure it is of the same type and on the
+                # same device as the data (compute follows data).
+                components = xp.asarray(components, device=X.device)
+                mean = xp.asarray(mean, device=X.device)
+
+            return X @ components + mean
+
+        def _onedal_supported(self, method_name, X):
+            class_name = self.__class__.__name__
+            patching_status = PatchingConditionsChain(
+                f"sklearn.decomposition.{class_name}.{method_name}"
+            )
+
+            if method_name == "fit":
+                shape_tuple, _is_shape_compatible = self._get_shape_compatibility(X)
+                patching_status.and_conditions(
+                    [
+                        (
+                            _is_shape_compatible,
+                            "Data shape is not compatible.",
+                        ),
+                        (
+                            self._is_solver_compatible_with_onedal(shape_tuple),
+                            (
+                                "Only 'covariance_eigh' and 'onedal_svd' "
+                                "solvers are supported."
+                                if sklearn_check_version("1.5")
+                                else "Only 'full', 'covariance_eigh' and 'onedal_svd' "
+                                "solvers are supported."
+                            ),
+                        ),
+                        (not issparse(X), "oneDAL PCA does not support sparse data"),
+                    ]
+                )
+                return patching_status
+
+            if method_name == "transform":
+                patching_status.and_conditions(
+                    [
+                        (
+                            hasattr(self, "_onedal_estimator"),
+                            "oneDAL model was not trained",
+                        ),
+                    ]
+                )
+                return patching_status
+
+            raise RuntimeError(
+                f"Unknown method {method_name} in {self.__class__.__name__}"
+            )
+
+        def _onedal_cpu_supported(self, method_name, *data):
+            return self._onedal_supported(method_name, *data)
+
+        def _onedal_gpu_supported(self, method_name, *data):
+            return self._onedal_supported(method_name, *data)
+
+        def _get_shape_compatibility(self, X):
+            _is_shape_compatible = False
+            _empty_shape = (0, 0)
+            if hasattr(X, "shape"):
+                shape_tuple = X.shape
+                if len(shape_tuple) == 1:
+                    shape_tuple = (1, shape_tuple[0])
+            elif isinstance(X, list):
+                if np.ndim(X) == 1:
+                    shape_tuple = (1, len(X))
+                elif np.ndim(X) == 2:
+                    shape_tuple = (len(X), len(X[0]))
+            else:
+                return _empty_shape, _is_shape_compatible
+
+            if shape_tuple[0] > 0 and shape_tuple[1] > 0 and len(shape_tuple) == 2:
+                _is_shape_compatible = shape_tuple[1] / shape_tuple[0] < 2
+
+            return shape_tuple, _is_shape_compatible
+
+        def _is_solver_compatible_with_onedal(self, shape_tuple):
+            self._fit_svd_solver = self.svd_solver
+            n_sf_min = min(shape_tuple)
+            n_components = n_sf_min if self.n_components is None else self.n_components
+
+            if self._fit_svd_solver == "auto":
+                if sklearn_check_version("1.1"):
+                    if (
+                        sklearn_check_version("1.5")
+                        and shape_tuple[1] <= 1_000
+                        and shape_tuple[0] >= 10 * shape_tuple[1]
+                    ):
+                        self._fit_svd_solver = "covariance_eigh"
+                    elif max(shape_tuple) <= 500 or n_components == "mle":
+                        self._fit_svd_solver = "full"
+                    elif 1 <= n_components < 0.8 * n_sf_min:
+                        self._fit_svd_solver = "randomized"
+                    else:
+                        self._fit_svd_solver = "full"
+                else:
+                    if n_components == "mle":
+                        self._fit_svd_solver = "full"
+                    else:
+                        # check if sklearnex is faster than randomized sklearn
+                        # Refer to daal4py
+                        regression_coefs = np.array(
+                            [
+                                [
+                                    9.779873e-11,
+                                    shape_tuple[0] * shape_tuple[1] * n_components,
+                                ],
+                                [
+                                    -1.122062e-11,
+                                    shape_tuple[0] * shape_tuple[1] * shape_tuple[1],
+                                ],
+                                [1.127905e-09, shape_tuple[0] ** 2],
+                            ]
+                        )
+                        if (
+                            n_components >= 1
+                            and np.dot(regression_coefs[:, 0], regression_coefs[:, 1])
+                            <= 0
+                        ):
+                            self._fit_svd_solver = "randomized"
+                        else:
+                            self._fit_svd_solver = "full"
+
+            # Use oneDAL in next cases:
+            # 1. oneDAL SVD solver is explicitly set
+            # 2. solver is set or dispatched to "covariance_eigh"
+            # 3. solver is set or dispatched to "full" and sklearn version < 1.5
+            # 4. solver is set to "auto" and dispatched to "full"
+            if self._fit_svd_solver in ["onedal_svd", "covariance_eigh"]:
+                return True
+            elif not sklearn_check_version("1.5") and self._fit_svd_solver == "full":
+                self._fit_svd_solver = "covariance_eigh"
+                return True
+            elif self.svd_solver == "auto" and self._fit_svd_solver == "full":
+                warn(
+                    "Sklearnex always uses `covariance_eigh` solver instead of `full` "
+                    "when `svd_solver` parameter is set to `auto` "
+                    "for performance purposes."
+                )
+                self._fit_svd_solver = "covariance_eigh"
+                return True
+            else:
+                return False
+
+        def _save_attributes(self):
+            self.n_samples_ = self._onedal_estimator.n_samples_
+            if sklearn_check_version("1.2"):
+                self.n_features_in_ = self._onedal_estimator.n_features_
+            else:
+                self.n_features_ = self._onedal_estimator.n_features_
+                self.n_features_in_ = self._onedal_estimator.n_features_
+            self.n_components_ = self._onedal_estimator.n_components_
+            self.components_ = self._onedal_estimator.components_
+            self.mean_ = self._onedal_estimator.mean_
+            self.singular_values_ = self._onedal_estimator.singular_values_
+            self.explained_variance_ = self._onedal_estimator.explained_variance_.ravel()
+            self.explained_variance_ratio_ = (
+                self._onedal_estimator.explained_variance_ratio_
+            )
+            self.noise_variance_ = self._onedal_estimator.noise_variance_
+
+        def _validate_n_features_in_after_fitting(self, X):
+            if sklearn_check_version("1.2"):
+                expected_n_features = self.n_features_in_
+            else:
+                expected_n_features = self.n_features_
+            if X.shape[1] != expected_n_features:
+                raise ValueError(
+                    (
+                        f"X has {X.shape[1]} features, "
+                        f"but PCA is expecting {expected_n_features} features as input"
+                    )
+                )
+
+        fit.__doc__ = sklearn_PCA.fit.__doc__
+        transform.__doc__ = sklearn_PCA.transform.__doc__
+        fit_transform.__doc__ = sklearn_PCA.fit_transform.__doc__
+        inverse_transform.__doc__ = sklearn_PCA.inverse_transform.__doc__
+
+else:
+    from daal4py.sklearn.decomposition import PCA
+
+    logging.warning(
+        "Sklearnex PCA requires oneDAL version >= 2024.1.0 but it was not found"
+    )

sklearnex/decomposition/tests/test_pca.py

@@ -0,0 +1,58 @@
+# ===============================================================================
+# Copyright 2023 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 numpy as np
+import pytest
+from numpy.testing import assert_allclose
+
+from daal4py.sklearn._utils import daal_check_version
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_sklearnex_import(dataframe, queue):
+    from sklearnex.decomposition import PCA
+
+    X = [[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]]
+    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    X_transformed_expected = [
+        [-1.38340578, -0.2935787],
+        [-2.22189802, 0.25133484],
+        [-3.6053038, -0.04224385],
+        [1.38340578, 0.2935787],
+        [2.22189802, -0.25133484],
+        [3.6053038, 0.04224385],
+    ]
+
+    pca = PCA(n_components=2, svd_solver="covariance_eigh")
+    pca.fit(X)
+    X_transformed = pca.transform(X)
+    X_fit_transformed = PCA(n_components=2, svd_solver="covariance_eigh").fit_transform(X)
+
+    if daal_check_version((2024, "P", 100)):
+        assert "sklearnex" in pca.__module__
+        assert hasattr(pca, "_onedal_estimator")
+    else:
+        assert "daal4py" in pca.__module__
+
+    tol = 1e-5 if _as_numpy(X_transformed).dtype == np.float32 else 1e-7
+    assert_allclose([6.30061232, 0.54980396], _as_numpy(pca.singular_values_))
+    assert_allclose(X_transformed_expected, _as_numpy(X_transformed), rtol=tol)
+    assert_allclose(X_transformed_expected, _as_numpy(X_fit_transformed), rtol=tol)