scikit-learn-intelex 2025.1.0__py312-none-manylinux_2_28_x86_64.whl

sklearnex/linear_model/logistic_regression.py

@@ -0,0 +1,413 @@
+# ===============================================================================
+# 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 logging
+from abc import ABC
+
+from daal4py.sklearn._utils import daal_check_version
+from daal4py.sklearn.linear_model.logistic_path import (
+    LogisticRegression as _daal4py_LogisticRegression,
+)
+
+if daal_check_version((2024, "P", 1)):
+    import numpy as np
+    from scipy.sparse import issparse
+    from sklearn.linear_model import LogisticRegression as _sklearn_LogisticRegression
+    from sklearn.metrics import accuracy_score
+    from sklearn.utils.multiclass import type_of_target
+    from sklearn.utils.validation import check_array, check_is_fitted, check_X_y
+
+    from daal4py.sklearn._n_jobs_support import control_n_jobs
+    from daal4py.sklearn._utils import sklearn_check_version
+    from daal4py.sklearn.linear_model.logistic_path import daal4py_fit, daal4py_predict
+    from onedal.linear_model import LogisticRegression as onedal_LogisticRegression
+    from onedal.utils import _num_samples
+
+    from .._config import get_config
+    from .._device_offload import dispatch, wrap_output_data
+    from .._utils import PatchingConditionsChain, get_patch_message
+
+    if sklearn_check_version("1.6"):
+        from sklearn.utils.validation import validate_data
+    else:
+        validate_data = _sklearn_LogisticRegression._validate_data
+
+    _sparsity_enabled = daal_check_version((2024, "P", 700))
+
+    class BaseLogisticRegression(ABC):
+        def _save_attributes(self):
+            assert hasattr(self, "_onedal_estimator")
+            self.classes_ = self._onedal_estimator.classes_
+            self.coef_ = self._onedal_estimator.coef_
+            self.intercept_ = self._onedal_estimator.intercept_
+            self.n_features_in_ = self._onedal_estimator.n_features_in_
+            self.n_iter_ = self._onedal_estimator.n_iter_
+
+    @control_n_jobs(
+        decorated_methods=[
+            "fit",
+            "predict",
+            "predict_proba",
+            "predict_log_proba",
+            "score",
+        ]
+    )
+    class LogisticRegression(_sklearn_LogisticRegression, BaseLogisticRegression):
+        __doc__ = _sklearn_LogisticRegression.__doc__
+
+        if sklearn_check_version("1.2"):
+            _parameter_constraints: dict = {
+                **_sklearn_LogisticRegression._parameter_constraints
+            }
+
+        def __init__(
+            self,
+            penalty="l2",
+            *,
+            dual=False,
+            tol=1e-4,
+            C=1.0,
+            fit_intercept=True,
+            intercept_scaling=1,
+            class_weight=None,
+            random_state=None,
+            solver="lbfgs",
+            max_iter=100,
+            multi_class="deprecated" if sklearn_check_version("1.5") else "auto",
+            verbose=0,
+            warm_start=False,
+            n_jobs=None,
+            l1_ratio=None,
+        ):
+            super().__init__(
+                penalty=penalty,
+                dual=dual,
+                tol=tol,
+                C=C,
+                fit_intercept=fit_intercept,
+                intercept_scaling=intercept_scaling,
+                class_weight=class_weight,
+                random_state=random_state,
+                solver=solver,
+                max_iter=max_iter,
+                multi_class=multi_class,
+                verbose=verbose,
+                warm_start=warm_start,
+                n_jobs=n_jobs,
+                l1_ratio=l1_ratio,
+            )
+
+        _onedal_cpu_fit = daal4py_fit
+
+        def fit(self, X, y, sample_weight=None):
+            if sklearn_check_version("1.2"):
+                self._validate_params()
+            dispatch(
+                self,
+                "fit",
+                {
+                    "onedal": self.__class__._onedal_fit,
+                    "sklearn": _sklearn_LogisticRegression.fit,
+                },
+                X,
+                y,
+                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_LogisticRegression.predict,
+                },
+                X,
+            )
+
+        @wrap_output_data
+        def predict_proba(self, X):
+            check_is_fitted(self)
+            return dispatch(
+                self,
+                "predict_proba",
+                {
+                    "onedal": self.__class__._onedal_predict_proba,
+                    "sklearn": _sklearn_LogisticRegression.predict_proba,
+                },
+                X,
+            )
+
+        @wrap_output_data
+        def predict_log_proba(self, X):
+            check_is_fitted(self)
+            return dispatch(
+                self,
+                "predict_log_proba",
+                {
+                    "onedal": self.__class__._onedal_predict_log_proba,
+                    "sklearn": _sklearn_LogisticRegression.predict_log_proba,
+                },
+                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_LogisticRegression.score,
+                },
+                X,
+                y,
+                sample_weight=sample_weight,
+            )
+
+        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
+            )
+
+        def _onedal_gpu_fit_supported(self, method_name, *data):
+            assert method_name == "fit"
+            assert len(data) == 3
+            X, y, sample_weight = data
+
+            class_name = self.__class__.__name__
+            patching_status = PatchingConditionsChain(
+                f"sklearn.linear_model.{class_name}.fit"
+            )
+
+            target_type = (
+                type_of_target(y, input_name="y")
+                if sklearn_check_version("1.1")
+                else type_of_target(y)
+            )
+            patching_status.and_conditions(
+                [
+                    (self.penalty == "l2", "Only l2 penalty is supported."),
+                    (self.dual == False, "dual=True is not supported."),
+                    (
+                        self.intercept_scaling == 1,
+                        "Intercept scaling is not supported.",
+                    ),
+                    (self.class_weight is None, "Class weight is not supported"),
+                    (self.solver == "newton-cg", "Only newton-cg solver is supported."),
+                    (
+                        self.multi_class != "multinomial",
+                        "multi_class parameter is not supported.",
+                    ),
+                    (self.warm_start == False, "Warm start is not supported."),
+                    (self.l1_ratio is None, "l1 ratio is not supported."),
+                    (sample_weight is None, "Sample weight is not supported."),
+                    (
+                        target_type == "binary",
+                        "Only binary classification is supported",
+                    ),
+                ]
+            )
+
+            return patching_status
+
+        def _onedal_gpu_predict_supported(self, method_name, *data):
+            assert method_name in [
+                "predict",
+                "predict_proba",
+                "predict_log_proba",
+                "score",
+            ]
+
+            class_name = self.__class__.__name__
+            patching_status = PatchingConditionsChain(
+                f"sklearn.linear_model.{class_name}.{method_name}"
+            )
+
+            n_samples = _num_samples(data[0])
+            dal_ready = patching_status.and_conditions(
+                [
+                    (n_samples > 0, "Number of samples is less than 1."),
+                    (
+                        (not any([issparse(i) for i in data])) or _sparsity_enabled,
+                        "Sparse input is not supported.",
+                    ),
+                    (
+                        hasattr(self, "_onedal_estimator"),
+                        "oneDAL model was not trained.",
+                    ),
+                ]
+            )
+
+            return patching_status
+
+        def _onedal_gpu_supported(self, method_name, *data):
+            if method_name == "fit":
+                return self._onedal_gpu_fit_supported(method_name, *data)
+            if method_name in ["predict", "predict_proba", "predict_log_proba", "score"]:
+                return self._onedal_gpu_predict_supported(method_name, *data)
+            raise RuntimeError(
+                f"Unknown method {method_name} in {self.__class__.__name__}"
+            )
+
+        def _onedal_cpu_supported(self, method_name, *data):
+            class_name = self.__class__.__name__
+            patching_status = PatchingConditionsChain(
+                f"sklearn.linear_model.{class_name}.{method_name}"
+            )
+
+            return patching_status
+
+        def _initialize_onedal_estimator(self):
+            onedal_params = {
+                "tol": self.tol,
+                "C": self.C,
+                "fit_intercept": self.fit_intercept,
+                "solver": self.solver,
+                "max_iter": self.max_iter,
+            }
+            self._onedal_estimator = onedal_LogisticRegression(**onedal_params)
+
+        def _onedal_fit(self, X, y, sample_weight=None, queue=None):
+            if queue is None or queue.sycl_device.is_cpu:
+                return self._onedal_cpu_fit(X, y, sample_weight)
+
+            assert sample_weight is None
+
+            if sklearn_check_version("1.0"):
+                X, y = validate_data(
+                    self,
+                    X,
+                    y,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+            else:
+                X, y = check_X_y(
+                    X,
+                    y,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+
+            self._initialize_onedal_estimator()
+            if get_config()["allow_sklearn_after_onedal"]:
+                try:
+                    self._onedal_estimator.fit(X, y, queue=queue)
+                    self._save_attributes()
+                except RuntimeError:
+                    logging.getLogger("sklearnex").info(
+                        f"{self.__class__.__name__}.fit "
+                        + get_patch_message("sklearn_after_onedal")
+                    )
+
+                    del self._onedal_estimator
+                    super().fit(X, y)
+            else:
+                self._onedal_estimator.fit(X, y, queue=queue)
+                self._save_attributes()
+
+        def _onedal_predict(self, X, queue=None):
+            if queue is None or queue.sycl_device.is_cpu:
+                return daal4py_predict(self, X, "computeClassLabels")
+
+            if sklearn_check_version("1.0"):
+                X = validate_data(
+                    self,
+                    X,
+                    reset=False,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+            else:
+                X = check_array(
+                    X,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+
+            assert hasattr(self, "_onedal_estimator")
+            return self._onedal_estimator.predict(X, queue=queue)
+
+        def _onedal_predict_proba(self, X, queue=None):
+            if queue is None or queue.sycl_device.is_cpu:
+                return daal4py_predict(self, X, "computeClassProbabilities")
+
+            if sklearn_check_version("1.0"):
+                X = validate_data(
+                    self,
+                    X,
+                    reset=False,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+            else:
+                X = check_array(
+                    X,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+
+            assert hasattr(self, "_onedal_estimator")
+            return self._onedal_estimator.predict_proba(X, queue=queue)
+
+        def _onedal_predict_log_proba(self, X, queue=None):
+            if queue is None or queue.sycl_device.is_cpu:
+                return daal4py_predict(self, X, "computeClassLogProbabilities")
+
+            if sklearn_check_version("1.0"):
+                X = validate_data(
+                    self,
+                    X,
+                    reset=False,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+            else:
+                X = check_array(
+                    X,
+                    accept_sparse=_sparsity_enabled,
+                    accept_large_sparse=_sparsity_enabled,
+                    dtype=[np.float64, np.float32],
+                )
+
+            assert hasattr(self, "_onedal_estimator")
+            return self._onedal_estimator.predict_log_proba(X, queue=queue)
+
+        fit.__doc__ = _sklearn_LogisticRegression.fit.__doc__
+        predict.__doc__ = _sklearn_LogisticRegression.predict.__doc__
+        predict_proba.__doc__ = _sklearn_LogisticRegression.predict_proba.__doc__
+        predict_log_proba.__doc__ = _sklearn_LogisticRegression.predict_log_proba.__doc__
+        score.__doc__ = _sklearn_LogisticRegression.score.__doc__
+
+else:
+    LogisticRegression = _daal4py_LogisticRegression
+
+    logging.warning(
+        "Sklearnex LogisticRegression requires oneDAL version >= 2024.0.1 "
+        "but it was not found"
+    )

sklearnex/linear_model/ridge.py

@@ -0,0 +1,24 @@
+# ===============================================================================
+# 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 daal4py.sklearn.linear_model import Ridge
+from onedal._device_offload import support_input_format
+
+# Note: `sklearnex.linear_model.Ridge` only has functional
+# sycl GPU support. No GPU device will be offloaded.
+Ridge.fit = support_input_format(queue_param=False)(Ridge.fit)
+Ridge.predict = support_input_format(queue_param=False)(Ridge.predict)
+Ridge.score = support_input_format(queue_param=False)(Ridge.score)

sklearnex/linear_model/tests/test_incremental_linear.py

@@ -0,0 +1,207 @@
+# ===============================================================================
+# 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 numpy as np
+import pytest
+from numpy.testing import assert_allclose
+
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex.linear_model import IncrementalLinearRegression
+from sklearnex.tests.utils import _IS_INTEL
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("fit_intercept", [True, False])
+@pytest.mark.parametrize("macro_block", [None, 1024])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_fit_on_gold_data(dataframe, queue, fit_intercept, macro_block, dtype):
+    X = np.array([[1], [2]])
+    X = X.astype(dtype=dtype)
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y = np.array([[1], [2]])
+    y = y.astype(dtype=dtype)
+    y_df = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+
+    inclin = IncrementalLinearRegression(fit_intercept=fit_intercept)
+    if macro_block is not None:
+        hparams = inclin.get_hyperparameters("fit")
+        hparams.cpu_macro_block = macro_block
+        hparams.gpu_macro_block = macro_block
+    inclin.fit(X_df, y_df)
+
+    y_pred = inclin.predict(X_df)
+    np_y_pred = _as_numpy(y_pred)
+
+    tol = 2e-6 if dtype == np.float32 else 1e-7
+    assert_allclose(inclin.coef_, [1], atol=tol)
+    if fit_intercept:
+        assert_allclose(inclin.intercept_, [0], atol=tol)
+    assert_allclose(np_y_pred, y, atol=tol)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("fit_intercept", [True, False])
+@pytest.mark.parametrize("macro_block", [None, 1024])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_partial_fit_on_gold_data(
+    dataframe, queue, fit_intercept, macro_block, dtype
+):
+    X = np.array([[1], [2], [3], [4]])
+    X = X.astype(dtype=dtype)
+    y = X + 3
+    y = y.astype(dtype=dtype)
+    X_split = np.array_split(X, 2)
+    y_split = np.array_split(y, 2)
+
+    inclin = IncrementalLinearRegression()
+    if macro_block is not None:
+        hparams = inclin.get_hyperparameters("fit")
+        hparams.cpu_macro_block = macro_block
+        hparams.gpu_macro_block = macro_block
+    for i in range(2):
+        X_split_df = _convert_to_dataframe(
+            X_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        y_split_df = _convert_to_dataframe(
+            y_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        inclin.partial_fit(X_split_df, y_split_df)
+
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_pred = inclin.predict(X_df)
+    np_y_pred = _as_numpy(y_pred)
+
+    assert inclin.n_features_in_ == 1
+    tol = 1e-5 if dtype == np.float32 else 1e-7
+    assert_allclose(inclin.coef_, [[1]], atol=tol)
+    if fit_intercept:
+        assert_allclose(inclin.intercept_, 3, atol=tol)
+
+    assert_allclose(np_y_pred, y, atol=tol)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("fit_intercept", [True, False])
+@pytest.mark.parametrize("macro_block", [None, 1024])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_partial_fit_multitarget_on_gold_data(
+    dataframe, queue, fit_intercept, macro_block, dtype
+):
+    X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
+    X = X.astype(dtype=dtype)
+    y = np.dot(X, [1, 2]) + 3
+    y = y.astype(dtype=dtype)
+    X_split = np.array_split(X, 2)
+    y_split = np.array_split(y, 2)
+
+    inclin = IncrementalLinearRegression()
+    if macro_block is not None:
+        hparams = inclin.get_hyperparameters("fit")
+        hparams.cpu_macro_block = macro_block
+        hparams.gpu_macro_block = macro_block
+    for i in range(2):
+        X_split_df = _convert_to_dataframe(
+            X_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        y_split_df = _convert_to_dataframe(
+            y_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        inclin.partial_fit(X_split_df, y_split_df)
+
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_pred = inclin.predict(X_df)
+    np_y_pred = _as_numpy(y_pred)
+
+    assert inclin.n_features_in_ == 2
+    tol = 1e-7
+    if dtype == np.float32:
+        tol = 7e-6 if _IS_INTEL else 2e-5
+
+    assert_allclose(inclin.coef_, [1.0, 2.0], atol=tol)
+    if fit_intercept:
+        assert_allclose(inclin.intercept_, 3.0, atol=tol)
+
+    assert_allclose(np_y_pred, y, atol=tol)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("fit_intercept", [True, False])
+@pytest.mark.parametrize("num_samples", [100, 1000])
+@pytest.mark.parametrize("num_features", [5, 10])
+@pytest.mark.parametrize("num_targets", [1, 2])
+@pytest.mark.parametrize("num_blocks", [1, 10])
+@pytest.mark.parametrize("macro_block", [None, 1024])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_partial_fit_on_random_data(
+    dataframe,
+    queue,
+    fit_intercept,
+    num_samples,
+    num_features,
+    num_targets,
+    num_blocks,
+    macro_block,
+    dtype,
+):
+    seed = 42
+    gen = np.random.default_rng(seed)
+    intercept = gen.random(size=num_targets, dtype=dtype)
+    coef = gen.random(size=(num_targets, num_features), dtype=dtype).T
+
+    X = gen.random(size=(num_samples, num_features), dtype=dtype)
+    if fit_intercept:
+        y = X @ coef + intercept[np.newaxis, :]
+    else:
+        y = X @ coef
+
+    X_split = np.array_split(X, num_blocks)
+    y_split = np.array_split(y, num_blocks)
+
+    inclin = IncrementalLinearRegression(fit_intercept=fit_intercept)
+    if macro_block is not None:
+        hparams = inclin.get_hyperparameters("fit")
+        hparams.cpu_macro_block = macro_block
+        hparams.gpu_macro_block = macro_block
+    for i in range(num_blocks):
+        X_split_df = _convert_to_dataframe(
+            X_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        y_split_df = _convert_to_dataframe(
+            y_split[i], sycl_queue=queue, target_df=dataframe
+        )
+        inclin.partial_fit(X_split_df, y_split_df)
+
+    tol = 1e-4 if inclin.coef_.dtype == np.float32 else 1e-7
+    assert_allclose(coef.T.squeeze(), inclin.coef_, atol=tol)
+
+    if fit_intercept:
+        assert_allclose(intercept, inclin.intercept_, atol=tol)
+
+    X_test = gen.random(size=(num_samples, num_features), dtype=dtype)
+    if fit_intercept:
+        expected_y_pred = (X_test @ coef + intercept[np.newaxis, :]).squeeze()
+    else:
+        expected_y_pred = (X_test @ coef).squeeze()
+
+    X_test_df = _convert_to_dataframe(X_test, sycl_queue=queue, target_df=dataframe)
+
+    y_pred = inclin.predict(X_test_df)
+
+    assert_allclose(expected_y_pred, _as_numpy(y_pred), atol=tol)