scikit-learn-intelex 2024.4.0__py310-none-manylinux1_x86_64.whl → 2024.6.0__py310-none-manylinux1_x86_64.whl

sklearnex/covariance/tests/test_incremental_covariance.py

@@ -17,6 +17,10 @@
 import numpy as np
 import pytest
 from numpy.testing import assert_allclose
+from sklearn.covariance.tests.test_covariance import (
+    test_covariance,
+    test_EmpiricalCovariance_validates_mahalanobis,
+)
 
 from onedal.tests.utils._dataframes_support import (
     _convert_to_dataframe,
@@ -26,13 +30,14 @@ from onedal.tests.utils._dataframes_support import (
 
 @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
 @pytest.mark.parametrize("dtype", [np.float32, np.float64])
-def test_sklearnex_partial_fit_on_gold_data(dataframe, queue, dtype):
+@pytest.mark.parametrize("assume_centered", [True, False])
+def test_sklearnex_partial_fit_on_gold_data(dataframe, queue, dtype, assume_centered):
     from sklearnex.covariance import IncrementalEmpiricalCovariance
 
     X = np.array([[0, 1], [0, 1]])
     X = X.astype(dtype)
     X_split = np.array_split(X, 2)
-    inccov = IncrementalEmpiricalCovariance()
+    inccov = IncrementalEmpiricalCovariance(assume_centered=assume_centered)
 
     for i in range(2):
         X_split_df = _convert_to_dataframe(
@@ -40,8 +45,12 @@ def test_sklearnex_partial_fit_on_gold_data(dataframe, queue, dtype):
         )
         result = inccov.partial_fit(X_split_df)
 
-    expected_covariance = np.array([[0, 0], [0, 0]])
-    expected_means = np.array([0, 1])
+    if assume_centered:
+        expected_covariance = np.array([[0, 0], [0, 1]])
+        expected_means = np.array([0, 0])
+    else:
+        expected_covariance = np.array([[0, 0], [0, 0]])
+        expected_means = np.array([0, 1])
 
     assert_allclose(expected_covariance, result.covariance_)
     assert_allclose(expected_means, result.location_)
@@ -49,7 +58,7 @@ def test_sklearnex_partial_fit_on_gold_data(dataframe, queue, dtype):
     X = np.array([[1, 2], [3, 6]])
     X = X.astype(dtype)
     X_split = np.array_split(X, 2)
-    inccov = IncrementalEmpiricalCovariance()
+    inccov = IncrementalEmpiricalCovariance(assume_centered=assume_centered)
 
     for i in range(2):
         X_split_df = _convert_to_dataframe(
@@ -57,8 +66,12 @@ def test_sklearnex_partial_fit_on_gold_data(dataframe, queue, dtype):
         )
         result = inccov.partial_fit(X_split_df)
 
-    expected_covariance = np.array([[1, 2], [2, 4]])
-    expected_means = np.array([2, 4])
+    if assume_centered:
+        expected_covariance = np.array([[5, 10], [10, 20]])
+        expected_means = np.array([0, 0])
+    else:
+        expected_covariance = np.array([[1, 2], [2, 4]])
+        expected_means = np.array([2, 4])
 
     assert_allclose(expected_covariance, result.covariance_)
     assert_allclose(expected_means, result.location_)
@@ -87,9 +100,9 @@ def test_sklearnex_fit_on_gold_data(dataframe, queue, batch_size, dtype):
 
 
 @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
-@pytest.mark.parametrize("num_batches", [2, 4, 6, 8, 10])
-@pytest.mark.parametrize("row_count", [100, 1000, 2000])
-@pytest.mark.parametrize("column_count", [10, 100, 200])
+@pytest.mark.parametrize("num_batches", [2, 10])
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
 @pytest.mark.parametrize("dtype", [np.float32, np.float64])
 def test_sklearnex_partial_fit_on_random_data(
     dataframe, queue, num_batches, row_count, column_count, dtype
@@ -117,12 +130,13 @@ def test_sklearnex_partial_fit_on_random_data(
 
 
 @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
-@pytest.mark.parametrize("num_batches", [2, 4, 6, 8, 10])
-@pytest.mark.parametrize("row_count", [100, 1000, 2000])
-@pytest.mark.parametrize("column_count", [10, 100, 200])
+@pytest.mark.parametrize("num_batches", [2, 10])
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
 @pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.parametrize("assume_centered", [True, False])
 def test_sklearnex_fit_on_random_data(
-    dataframe, queue, num_batches, row_count, column_count, dtype
+    dataframe, queue, num_batches, row_count, column_count, dtype, assume_centered
 ):
     from sklearnex.covariance import IncrementalEmpiricalCovariance
 
@@ -132,12 +146,35 @@ def test_sklearnex_fit_on_random_data(
     X = X.astype(dtype)
     X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
     batch_size = row_count // num_batches
-    inccov = IncrementalEmpiricalCovariance(batch_size=batch_size)
+    inccov = IncrementalEmpiricalCovariance(
+        batch_size=batch_size, assume_centered=assume_centered
+    )
 
     result = inccov.fit(X_df)
 
-    expected_covariance = np.cov(X.T, bias=1)
-    expected_means = np.mean(X, axis=0)
+    if assume_centered:
+        expected_covariance = np.dot(X.T, X) / X.shape[0]
+        expected_means = np.zeros_like(X[0])
+    else:
+        expected_covariance = np.cov(X.T, bias=1)
+        expected_means = np.mean(X, axis=0)
 
     assert_allclose(expected_covariance, result.covariance_, atol=1e-6)
     assert_allclose(expected_means, result.location_, atol=1e-6)
+
+
+# Monkeypatch IncrementalEmpiricalCovariance into relevant sklearn.covariance tests
+@pytest.mark.allow_sklearn_fallback
+@pytest.mark.parametrize(
+    "sklearn_test",
+    [
+        test_covariance,
+        test_EmpiricalCovariance_validates_mahalanobis,
+    ],
+)
+def test_IncrementalEmpiricalCovariance_against_sklearn(monkeypatch, sklearn_test):
+    from sklearnex.covariance import IncrementalEmpiricalCovariance
+
+    class_name = ".".join([sklearn_test.__module__, "EmpiricalCovariance"])
+    monkeypatch.setattr(class_name, IncrementalEmpiricalCovariance)
+    sklearn_test()

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

@@ -41,16 +41,18 @@ 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__
         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))
-    assert_allclose(X_transformed_expected, _as_numpy(X_fit_transformed))
+    assert_allclose(X_transformed_expected, _as_numpy(X_transformed), rtol=tol)
+    assert_allclose(X_transformed_expected, _as_numpy(X_fit_transformed), rtol=tol)

sklearnex/dispatcher.py

@@ -45,12 +45,14 @@ def get_patch_map_core(preview=False):
 
         if _is_new_patching_available():
             import sklearn.covariance as covariance_module
+            import sklearn.decomposition as decomposition_module
 
             # Preview classes for patching
             from .preview.cluster import KMeans as KMeans_sklearnex
             from .preview.covariance import (
                 EmpiricalCovariance as EmpiricalCovariance_sklearnex,
             )
+            from .preview.decomposition import IncrementalPCA as IncrementalPCA_sklearnex
 
             # Since the state of the lru_cache without preview cannot be
             # guaranteed to not have already enabled sklearnex algorithms
@@ -62,7 +64,7 @@ def get_patch_map_core(preview=False):
             sklearn_obj = mapping["kmeans"][0][1]
             mapping.pop("kmeans")
             mapping["kmeans"] = [
-                [(cluster_module, "kmeans", KMeans_sklearnex), sklearn_obj]
+                [(cluster_module, "KMeans", KMeans_sklearnex), sklearn_obj]
             ]
 
             # Covariance
@@ -76,6 +78,18 @@ def get_patch_map_core(preview=False):
                     None,
                 ]
             ]
+
+            # IncrementalPCA
+            mapping["incrementalpca"] = [
+                [
+                    (
+                        decomposition_module,
+                        "IncrementalPCA",
+                        IncrementalPCA_sklearnex,
+                    ),
+                    None,
+                ]
+            ]
         return mapping
 
     from daal4py.sklearn.monkeypatch.dispatcher import _get_map_of_algorithms
@@ -93,6 +107,7 @@ 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 +130,17 @@ 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 +294,30 @@ 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]