scikit-learn-intelex 2024.4.0__py39-none-manylinux1_x86_64.whl → 2024.5.0__py39-none-manylinux1_x86_64.whl

sklearnex/decomposition/pca.py

@@ -21,6 +21,7 @@ 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
@@ -35,9 +36,13 @@ if daal_check_version((2024, "P", 100)):
     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
+    from sklearnex.utils import get_namespace
 
     @control_n_jobs(decorated_methods=["fit", "transform", "fit_transform"])
     class PCA(sklearn_PCA):
@@ -45,6 +50,16 @@ if daal_check_version((2024, "P", 100)):
 
         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"):
 
@@ -107,7 +122,7 @@ if daal_check_version((2024, "P", 100)):
                     target_type=numbers.Integral,
                 )
 
-            U, S, Vt = dispatch(
+            return dispatch(
                 self,
                 "fit",
                 {
@@ -116,7 +131,6 @@ if daal_check_version((2024, "P", 100)):
                 },
                 X,
             )
-            return U, S, Vt
 
         def _onedal_fit(self, X, queue=None):
             X = self._validate_data(
@@ -129,7 +143,7 @@ if daal_check_version((2024, "P", 100)):
             onedal_params = {
                 "n_components": self.n_components,
                 "is_deterministic": True,
-                "method": "cov",
+                "method": "svd" if self._fit_svd_solver == "onedal_svd" else "cov",
                 "whiten": self.whiten,
             }
             self._onedal_estimator = onedal_PCA(**onedal_params)
@@ -140,7 +154,13 @@ if daal_check_version((2024, "P", 100)):
             S = self.singular_values_
             Vt = self.components_
 
-            return U, S, Vt
+            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):
@@ -156,32 +176,39 @@ if daal_check_version((2024, "P", 100)):
 
         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)
-            if sklearn_check_version("1.0"):
-                self._check_feature_names(X, reset=False)
 
             return self._onedal_estimator.predict(X, queue=queue)
 
         def fit_transform(self, X, y=None):
-            U, S, Vt = self._fit(X)
-            if U is 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)
-            else:
+            elif U is not None:
                 # Scikit-learn PCA was fit
                 U = U[:, : self.n_components_]
 
                 if self.whiten:
-                    U *= sqrt(X.shape[0] - 1)
+                    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)
 
         def _onedal_supported(self, method_name, X):
             class_name = self.__class__.__name__
@@ -199,7 +226,13 @@ if daal_check_version((2024, "P", 100)):
                         ),
                         (
                             self._is_solver_compatible_with_onedal(shape_tuple),
-                            f"Only 'full' svd solver is supported.",
+                            (
+                                "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"),
                     ]
@@ -254,7 +287,13 @@ if daal_check_version((2024, "P", 100)):
 
             if self._fit_svd_solver == "auto":
                 if sklearn_check_version("1.1"):
-                    if max(shape_tuple) <= 500 or n_components == "mle":
+                    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"
@@ -288,7 +327,23 @@ if daal_check_version((2024, "P", 100)):
                         else:
                             self._fit_svd_solver = "full"
 
-            if 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

sklearnex/decomposition/tests/test_pca.py

@@ -41,10 +41,10 @@ def test_sklearnex_import(dataframe, queue):
         [3.6053038, 0.04224385],
     ]
 
-    pca = PCA(n_components=2, svd_solver="full")
+    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="full").fit_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__

sklearnex/dispatcher.py

@@ -93,6 +93,7 @@ def get_patch_map_core(preview=False):
         # Scikit-learn* modules
         import sklearn as base_module
         import sklearn.cluster as cluster_module
+        import sklearn.covariance as covariance_module
         import sklearn.decomposition as decomposition_module
         import sklearn.ensemble as ensemble_module
         import sklearn.linear_model as linear_model_module
@@ -115,11 +116,17 @@ def get_patch_map_core(preview=False):
             from .utils.parallel import _FuncWrapperOld as _FuncWrapper_sklearnex
 
         from .cluster import DBSCAN as DBSCAN_sklearnex
+        from .covariance import (
+            IncrementalEmpiricalCovariance as IncrementalEmpiricalCovariance_sklearnex,
+        )
         from .decomposition import PCA as PCA_sklearnex
         from .ensemble import ExtraTreesClassifier as ExtraTreesClassifier_sklearnex
         from .ensemble import ExtraTreesRegressor as ExtraTreesRegressor_sklearnex
         from .ensemble import RandomForestClassifier as RandomForestClassifier_sklearnex
         from .ensemble import RandomForestRegressor as RandomForestRegressor_sklearnex
+        from .linear_model import (
+            IncrementalLinearRegression as IncrementalLinearRegression_sklearnex,
+        )
         from .linear_model import LinearRegression as LinearRegression_sklearnex
         from .linear_model import LogisticRegression as LogisticRegression_sklearnex
         from .neighbors import KNeighborsClassifier as KNeighborsClassifier_sklearnex
@@ -273,6 +280,30 @@ def get_patch_map_core(preview=False):
         ]
         mapping["localoutlierfactor"] = mapping["lof"]
 
+        # IncrementalEmpiricalCovariance
+        mapping["incrementalempiricalcovariance"] = [
+            [
+                (
+                    covariance_module,
+                    "IncrementalEmpiricalCovariance",
+                    IncrementalEmpiricalCovariance_sklearnex,
+                ),
+                None,
+            ]
+        ]
+
+        # IncrementalLinearRegression
+        mapping["incrementallinearregression"] = [
+            [
+                (
+                    linear_model_module,
+                    "IncrementalLinearRegression",
+                    IncrementalLinearRegression_sklearnex,
+                ),
+                None,
+            ]
+        ]
+
         # Configs
         mapping["set_config"] = [
             [(base_module, "set_config", set_config_sklearnex), None]

sklearnex/ensemble/_forest.py

@@ -777,15 +777,16 @@ class ForestClassifier(sklearn_ForestClassifier, BaseForest):
         return patching_status
 
     def _onedal_predict(self, X, queue=None):
+        check_is_fitted(self, "_onedal_estimator")
+
+        if sklearn_check_version("1.0"):
+            self._check_feature_names(X, reset=False)
+
         X = check_array(
             X,
             dtype=[np.float64, np.float32],
             force_all_finite=False,
         )  # Warning, order of dtype matters
-        check_is_fitted(self, "_onedal_estimator")
-
-        if sklearn_check_version("1.0"):
-            self._check_feature_names(X, reset=False)
 
         res = self._onedal_estimator.predict(X, queue=queue)
         return np.take(self.classes_, res.ravel().astype(np.int64, casting="unsafe"))

sklearnex/linear_model/__init__.py

@@ -15,14 +15,16 @@
 # ===============================================================================
 
 from .coordinate_descent import ElasticNet, Lasso
+from .incremental_linear import IncrementalLinearRegression
 from .linear import LinearRegression
 from .logistic_regression import LogisticRegression
 from .ridge import Ridge
 
 __all__ = [
-    "Ridge",
-    "LinearRegression",
-    "LogisticRegression",
     "ElasticNet",
+    "IncrementalLinearRegression",
     "Lasso",
+    "LinearRegression",
+    "LogisticRegression",
+    "Ridge",
 ]

sklearnex/linear_model/incremental_linear.py

@@ -0,0 +1,387 @@
+# ===============================================================================
+# 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 numbers
+import warnings
+
+import numpy as np
+from sklearn.base import BaseEstimator, MultiOutputMixin, RegressorMixin
+from sklearn.exceptions import NotFittedError
+from sklearn.utils import check_array, gen_batches
+
+from daal4py.sklearn._n_jobs_support import control_n_jobs
+from daal4py.sklearn._utils import sklearn_check_version
+from onedal.linear_model import (
+    IncrementalLinearRegression as onedal_IncrementalLinearRegression,
+)
+
+if sklearn_check_version("1.2"):
+    from sklearn.utils._param_validation import Interval
+
+from onedal.common.hyperparameters import get_hyperparameters
+
+from .._device_offload import dispatch, wrap_output_data
+from .._utils import PatchingConditionsChain, register_hyperparameters
+
+
+@register_hyperparameters(
+    {
+        "fit": get_hyperparameters("linear_regression", "train"),
+        "partial_fit": get_hyperparameters("linear_regression", "train"),
+    }
+)
+@control_n_jobs(
+    decorated_methods=["fit", "partial_fit", "predict", "_onedal_finalize_fit"]
+)
+class IncrementalLinearRegression(MultiOutputMixin, RegressorMixin, BaseEstimator):
+    """
+    Incremental estimator for linear regression.
+    Allows to train linear regression if data are splitted into batches.
+
+    Parameters
+    ----------
+    fit_intercept : bool, default=True
+    Whether to calculate the intercept for this model. If set
+    to False, no intercept will be used in calculations
+    (i.e. data is expected to be centered).
+
+    copy_X : bool, default=True
+        If True, X will be copied; else, it may be overwritten.
+
+    n_jobs : int, default=None
+        The number of jobs to use for the computation.
+
+    batch_size : int, default=None
+        The number of samples to use for each batch. Only used when calling
+        ``fit``. If ``batch_size`` is ``None``, then ``batch_size``
+        is inferred from the data and set to ``5 * n_features``, to provide a
+        balance between approximation accuracy and memory consumption.
+
+    Attributes
+    ----------
+    coef_ : array of shape (n_features, ) or (n_targets, n_features)
+        Estimated coefficients for the linear regression problem.
+        If multiple targets are passed during the fit (y 2D), this
+        is a 2D array of shape (n_targets, n_features), while if only
+        one target is passed, this is a 1D array of length n_features.
+
+    intercept_ : float or array of shape (n_targets,)
+        Independent term in the linear model. Set to 0.0 if
+        `fit_intercept = False`.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit`.
+
+    n_samples_seen_ : int
+        The number of samples processed by the estimator. Will be reset on
+        new calls to fit, but increments across ``partial_fit`` calls.
+        It should be not less than `n_features_in_` if `fit_intercept`
+        is False and not less than `n_features_in_` + 1 if `fit_intercept`
+        is True to obtain regression coefficients.
+
+    batch_size_ : int
+        Inferred batch size from ``batch_size``.
+
+    n_features_in_ : int
+        Number of features seen during :term:`fit` `partial_fit`.
+
+    """
+
+    _onedal_incremental_linear = staticmethod(onedal_IncrementalLinearRegression)
+
+    if sklearn_check_version("1.2"):
+        _parameter_constraints: dict = {
+            "fit_intercept": ["boolean"],
+            "copy_X": ["boolean"],
+            "n_jobs": [Interval(numbers.Integral, -1, None, closed="left"), None],
+            "batch_size": [Interval(numbers.Integral, 1, None, closed="left"), None],
+        }
+
+    def __init__(self, *, fit_intercept=True, copy_X=True, n_jobs=None, batch_size=None):
+        self.fit_intercept = fit_intercept
+        self.copy_X = copy_X
+        self.n_jobs = n_jobs
+        self.batch_size = batch_size
+
+    def _onedal_supported(self, method_name, *data):
+        patching_status = PatchingConditionsChain(
+            f"sklearn.linear_model.{self.__class__.__name__}.{method_name}"
+        )
+        return patching_status
+
+    _onedal_cpu_supported = _onedal_supported
+    _onedal_gpu_supported = _onedal_supported
+
+    def _onedal_predict(self, X, queue=None):
+        if sklearn_check_version("1.2"):
+            self._validate_params()
+
+        if sklearn_check_version("1.0"):
+            X = self._validate_data(
+                X,
+                dtype=[np.float64, np.float32],
+                copy=self.copy_X,
+            )
+        else:
+            X = check_array(
+                X,
+                dtype=[np.float64, np.float32],
+                copy=self.copy_X,
+            )
+
+        assert hasattr(self, "_onedal_estimator")
+        if self._need_to_finalize:
+            self._onedal_finalize_fit()
+        return self._onedal_estimator.predict(X, queue)
+
+    def _onedal_partial_fit(self, X, y, queue=None):
+        first_pass = not hasattr(self, "n_samples_seen_") or self.n_samples_seen_ == 0
+
+        if sklearn_check_version("1.2"):
+            self._validate_params()
+
+        if sklearn_check_version("1.0"):
+            X, y = self._validate_data(
+                X,
+                y,
+                dtype=[np.float64, np.float32],
+                reset=first_pass,
+                copy=self.copy_X,
+                multi_output=True,
+            )
+        else:
+            X = check_array(
+                X,
+                dtype=[np.float64, np.float32],
+                copy=self.copy_X,
+            )
+            y = check_array(
+                y,
+                dtype=[np.float64, np.float32],
+                copy=False,
+                ensure_2d=False,
+            )
+
+        if first_pass:
+            self.n_samples_seen_ = X.shape[0]
+            self.n_features_in_ = X.shape[1]
+        else:
+            self.n_samples_seen_ += X.shape[0]
+        onedal_params = {"fit_intercept": self.fit_intercept, "copy_X": self.copy_X}
+        if not hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator = self._onedal_incremental_linear(**onedal_params)
+        self._onedal_estimator.partial_fit(X, y, queue)
+        self._need_to_finalize = True
+
+    def _onedal_finalize_fit(self):
+        assert hasattr(self, "_onedal_estimator")
+        is_underdetermined = self.n_samples_seen_ < self.n_features_in_ + int(
+            self.fit_intercept
+        )
+        if is_underdetermined:
+            raise ValueError("Not enough samples to finalize")
+        self._onedal_estimator.finalize_fit()
+        self._need_to_finalize = False
+
+    def _onedal_fit(self, X, y, queue=None):
+        if sklearn_check_version("1.2"):
+            self._validate_params()
+
+        if sklearn_check_version("1.0"):
+            X, y = self._validate_data(
+                X, y, dtype=[np.float64, np.float32], copy=self.copy_X, multi_output=True
+            )
+        else:
+            X = check_array(
+                X,
+                dtype=[np.float64, np.float32],
+                copy=self.copy_X,
+            )
+            y = check_array(
+                y,
+                dtype=[np.float64, np.float32],
+                copy=False,
+                ensure_2d=False,
+            )
+
+        n_samples, n_features = X.shape
+
+        is_underdetermined = n_samples < n_features + int(self.fit_intercept)
+        if is_underdetermined:
+            raise ValueError("Not enough samples to run oneDAL backend")
+
+        if self.batch_size is None:
+            self.batch_size_ = 5 * n_features
+        else:
+            self.batch_size_ = self.batch_size
+
+        self.n_samples_seen_ = 0
+        if hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator._reset()
+
+        for batch in gen_batches(n_samples, self.batch_size_):
+            X_batch, y_batch = X[batch], y[batch]
+            self._onedal_partial_fit(X_batch, y_batch, queue=queue)
+
+        if sklearn_check_version("1.2"):
+            self._validate_params()
+
+        # finite check occurs on onedal side
+        self.n_features_in_ = n_features
+
+        if n_samples == 1:
+            warnings.warn(
+                "Only one sample available. You may want to reshape your data array"
+            )
+
+        self._onedal_finalize_fit()
+
+        return self
+
+    def get_intercept_(self):
+        if hasattr(self, "_onedal_estimator"):
+            if self._need_to_finalize:
+                self._onedal_finalize_fit()
+
+            return self._onedal_estimator.intercept_
+        else:
+            raise AttributeError(
+                f"'{self.__class__.__name__}' object has no attribute 'intercept_'"
+            )
+
+    def set_intercept_(self, value):
+        self.__dict__["intercept_"] = value
+        if hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator.intercept_ = value
+            del self._onedal_estimator._onedal_model
+
+    def get_coef_(self):
+        if hasattr(self, "_onedal_estimator"):
+            if self._need_to_finalize:
+                self._onedal_finalize_fit()
+
+            return self._onedal_estimator.coef_
+        else:
+            raise AttributeError(
+                f"'{self.__class__.__name__}' object has no attribute 'coef_'"
+            )
+
+    def set_coef_(self, value):
+        self.__dict__["coef_"] = value
+        if hasattr(self, "_onedal_estimator"):
+            self._onedal_estimator.coef_ = value
+            del self._onedal_estimator._onedal_model
+
+    coef_ = property(get_coef_, set_coef_)
+    intercept_ = property(get_intercept_, set_intercept_)
+
+    def partial_fit(self, X, y):
+        """
+        Incremental fit linear model with X and y. All of X and y is
+        processed as a single batch.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_features)
+            Training data, where `n_samples` is the number of samples and
+            `n_features` is the number of features.
+
+        y : array-like of shape (n_samples,) or (n_samples, n_targets)
+            Target values, where `n_samples` is the number of samples and
+            `n_targets` is the number of targets.
+
+        Returns
+        -------
+        self : object
+            Returns the instance itself.
+        """
+
+        dispatch(
+            self,
+            "partial_fit",
+            {
+                "onedal": self.__class__._onedal_partial_fit,
+                "sklearn": None,
+            },
+            X,
+            y,
+        )
+        return self
+
+    def fit(self, X, y):
+        """
+        Fit the model with X and y, using minibatches of size batch_size.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_features)
+            Training data, where `n_samples` is the number of samples and
+            `n_features` is the number of features. It is necessary for
+            `n_samples` to be not less than `n_features` if `fit_intercept`
+            is False and not less than `n_features` + 1 if `fit_intercept`
+            is True
+
+        y : array-like of shape (n_samples,) or (n_samples, n_targets)
+            Target values, where `n_samples` is the number of samples and
+            `n_targets` is the number of targets.
+
+        Returns
+        -------
+        self : object
+            Returns the instance itself.
+        """
+
+        dispatch(
+            self,
+            "fit",
+            {
+                "onedal": self.__class__._onedal_fit,
+                "sklearn": None,
+            },
+            X,
+            y,
+        )
+        return self
+
+    @wrap_output_data
+    def predict(self, X, y=None):
+        """
+        Predict using the linear model.
+        Parameters
+        ----------
+        X : array-like or sparse matrix, shape (n_samples, n_features)
+            Samples.
+        Returns
+        -------
+        C : array, shape (n_samples, n_targets)
+            Returns predicted values.
+        """
+        if not hasattr(self, "coef_"):
+            msg = (
+                "This %(name)s instance is not fitted yet. Call 'fit' or 'partial_fit' "
+                "with appropriate arguments before using this estimator."
+            )
+            raise NotFittedError(msg % {"name": self.__class__.__name__})
+
+        return dispatch(
+            self,
+            "predict",
+            {
+                "onedal": self.__class__._onedal_predict,
+                "sklearn": None,
+            },
+            X,
+        )

sklearnex/linear_model/linear.py

@@ -157,7 +157,7 @@ class LinearRegression(sklearn_LinearRegression):
         n_features = _num_features(X, fallback_1d=True)
 
         # Check if equations are well defined
-        is_good_for_onedal = n_samples >= (n_features + int(self.fit_intercept))
+        is_underdetermined = n_samples < (n_features + int(self.fit_intercept))
 
         dal_ready = patching_status.and_conditions(
             [
@@ -172,7 +172,7 @@ class LinearRegression(sklearn_LinearRegression):
                     "Forced positive coefficients are not supported.",
                 ),
                 (
-                    is_good_for_onedal,
+                    not is_underdetermined,
                     "The shape of X (fitting) does not satisfy oneDAL requirements:"
                     "Number of features + 1 >= number of samples.",
                 ),