scikit-learn-intelex 2025.4.0__py313-none-manylinux_2_28_x86_64.whl

sklearnex/linear_model/tests/test_incremental_ridge.py

@@ -0,0 +1,214 @@
+# ===============================================================================
+# 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.
+# ===============================================================================
+
+from daal4py.sklearn._utils import daal_check_version
+
+if daal_check_version((2024, "P", 600)):
+    import numpy as np
+    import pytest
+    from numpy.testing import assert_allclose
+    from sklearn.exceptions import NotFittedError
+
+    from onedal.tests.utils._dataframes_support import (
+        _as_numpy,
+        _convert_to_dataframe,
+        get_dataframes_and_queues,
+    )
+    from sklearnex.linear_model import IncrementalRidge
+
+    def _compute_ridge_coefficients(X, y, alpha, fit_intercept):
+        coefficients_manual, intercept_manual = None, None
+        if fit_intercept:
+            X_mean = np.mean(X, axis=0)
+            y_mean = np.mean(y)
+            X_centered = X - X_mean
+            y_centered = y - y_mean
+
+            X_with_intercept = np.hstack([np.ones((X.shape[0], 1)), X_centered])
+            lambda_identity = alpha * np.eye(X_with_intercept.shape[1])
+            inverse_term = np.linalg.inv(
+                np.dot(X_with_intercept.T, X_with_intercept) + lambda_identity
+            )
+            xt_y = np.dot(X_with_intercept.T, y_centered)
+            coefficients_manual = np.dot(inverse_term, xt_y)
+
+            intercept_manual = y_mean - np.dot(X_mean, coefficients_manual[1:])
+            coefficients_manual = coefficients_manual[1:]
+        else:
+            lambda_identity = alpha * np.eye(X.shape[1])
+            inverse_term = np.linalg.inv(np.dot(X.T, X) + lambda_identity)
+            xt_y = np.dot(X.T, y)
+            coefficients_manual = np.dot(inverse_term, xt_y)
+
+        return coefficients_manual, intercept_manual
+
+    @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+    @pytest.mark.parametrize("batch_size", [10, 100, 1000])
+    @pytest.mark.parametrize("alpha", [0.1, 0.5, 1.0])
+    @pytest.mark.parametrize("fit_intercept", [True, False])
+    def test_inc_ridge_fit_coefficients(
+        dataframe, queue, alpha, batch_size, fit_intercept
+    ):
+        sample_size, feature_size = 1000, 50
+        X = np.random.rand(sample_size, feature_size)
+        y = np.random.rand(sample_size)
+        X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+        y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+
+        inc_ridge = IncrementalRidge(
+            fit_intercept=fit_intercept, alpha=alpha, batch_size=batch_size
+        )
+        inc_ridge.fit(X_c, y_c)
+
+        coefficients_manual, intercept_manual = _compute_ridge_coefficients(
+            X, y, alpha, fit_intercept
+        )
+        if fit_intercept:
+            assert_allclose(inc_ridge.intercept_, intercept_manual, rtol=1e-6, atol=1e-6)
+
+        assert_allclose(inc_ridge.coef_, coefficients_manual, rtol=1e-6, atol=1e-6)
+
+    @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+    @pytest.mark.parametrize("batch_size", [2, 5])
+    @pytest.mark.parametrize("alpha", [0.1, 0.5, 1.0])
+    def test_inc_ridge_partial_fit_coefficients(dataframe, queue, alpha, batch_size):
+        sample_size, feature_size = 1000, 50
+        X = np.random.rand(sample_size, feature_size)
+        y = np.random.rand(sample_size)
+        X_split = np.array_split(X, batch_size)
+        y_split = np.array_split(y, batch_size)
+
+        inc_ridge = IncrementalRidge(fit_intercept=False, alpha=alpha)
+
+        for batch_index in range(len(X_split)):
+            X_c = _convert_to_dataframe(
+                X_split[batch_index], sycl_queue=queue, target_df=dataframe
+            )
+            y_c = _convert_to_dataframe(
+                y_split[batch_index], sycl_queue=queue, target_df=dataframe
+            )
+            inc_ridge.partial_fit(X_c, y_c)
+
+        lambda_identity = alpha * np.eye(X.shape[1])
+        inverse_term = np.linalg.inv(np.dot(X.T, X) + lambda_identity)
+        xt_y = np.dot(X.T, y)
+        coefficients_manual = np.dot(inverse_term, xt_y)
+
+        assert_allclose(inc_ridge.coef_, coefficients_manual, rtol=1e-6, atol=1e-6)
+
+    def test_inc_ridge_score_before_fit():
+        X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
+        y = np.dot(X, np.array([1, 2])) + 3
+        inc_ridge = IncrementalRidge(alpha=0.5)
+        with pytest.raises(NotFittedError):
+            inc_ridge.score(X, y)
+
+    def test_inc_ridge_predict_before_fit():
+        X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
+        inc_ridge = IncrementalRidge(alpha=0.5)
+        with pytest.raises(NotFittedError):
+            inc_ridge.predict(X)
+
+    def test_inc_ridge_score_after_fit():
+        X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
+        y = np.dot(X, np.array([1, 2])) + 3
+        inc_ridge = IncrementalRidge(alpha=0.5)
+        inc_ridge.fit(X, y)
+        assert inc_ridge.score(X, y) >= 0.97
+
+    @pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+    @pytest.mark.parametrize("fit_intercept", [True, False])
+    def test_inc_ridge_predict_after_fit(dataframe, queue, fit_intercept):
+        sample_size, feature_size = 1000, 50
+        X = np.random.rand(sample_size, feature_size)
+        y = np.random.rand(sample_size)
+        X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+        y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+
+        inc_ridge = IncrementalRidge(fit_intercept=fit_intercept, alpha=0.5)
+        inc_ridge.fit(X_c, y_c)
+
+        y_pred = inc_ridge.predict(X_c)
+
+        coefficients_manual, intercept_manual = _compute_ridge_coefficients(
+            X, y, 0.5, fit_intercept
+        )
+        y_pred_manual = np.dot(X, coefficients_manual)
+        if fit_intercept:
+            y_pred_manual += intercept_manual
+
+        assert_allclose(_as_numpy(y_pred), y_pred_manual, rtol=1e-6, atol=1e-6)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("fit_intercept", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_incremental_estimatior_pickle(dataframe, queue, fit_intercept, dtype):
+    import pickle
+
+    from sklearnex.linear_model import IncrementalRidge
+
+    inc_ridge = IncrementalRidge()
+
+    # Check that estimator can be serialized without any data.
+    dump = pickle.dumps(inc_ridge)
+    inc_ridge_loaded = pickle.loads(dump)
+
+    seed = 77
+    gen = np.random.default_rng(seed)
+    intercept = gen.random(size=1, dtype=dtype)
+    coef = gen.random(size=(1, 10), dtype=dtype).T
+    X = gen.uniform(low=-0.3, high=+0.7, size=(30, 10))
+    X = X.astype(dtype)
+    if fit_intercept:
+        y = X @ coef + intercept[np.newaxis, :]
+    else:
+        y = X @ coef
+    X_split = np.array_split(X, 2)
+    y_split = np.array_split(y, 2)
+    X_split_df = _convert_to_dataframe(X_split[0], sycl_queue=queue, target_df=dataframe)
+    y_split_df = _convert_to_dataframe(y_split[0], sycl_queue=queue, target_df=dataframe)
+    inc_ridge.partial_fit(X_split_df, y_split_df)
+    inc_ridge_loaded.partial_fit(X_split_df, y_split_df)
+
+    # Check that estimator can be serialized after partial_fit call.
+    dump = pickle.dumps(inc_ridge_loaded)
+    inc_ridge_loaded = pickle.loads(dump)
+
+    assert inc_ridge.batch_size == inc_ridge_loaded.batch_size
+    assert inc_ridge.n_features_in_ == inc_ridge_loaded.n_features_in_
+    assert inc_ridge.n_samples_seen_ == inc_ridge_loaded.n_samples_seen_
+    assert inc_ridge.alpha == inc_ridge_loaded.alpha
+    if hasattr(inc_ridge, "_parameter_constraints"):
+        assert inc_ridge._parameter_constraints == inc_ridge_loaded._parameter_constraints
+    assert inc_ridge.n_jobs == inc_ridge_loaded.n_jobs
+
+    X_split_df = _convert_to_dataframe(X_split[1], sycl_queue=queue, target_df=dataframe)
+    y_split_df = _convert_to_dataframe(y_split[1], sycl_queue=queue, target_df=dataframe)
+    inc_ridge.partial_fit(X_split_df, y_split_df)
+    inc_ridge_loaded.partial_fit(X_split_df, y_split_df)
+    dump = pickle.dumps(inc_ridge)
+    inc_ridge_loaded = pickle.loads(dump)
+
+    assert_allclose(inc_ridge.coef_, inc_ridge_loaded.coef_, atol=1e-6)
+    assert_allclose(inc_ridge.intercept_, inc_ridge_loaded.intercept_, atol=1e-6)
+
+    # Check that finalized estimator can be serialized.
+    dump = pickle.dumps(inc_ridge_loaded)
+    inc_ridge_loaded = pickle.loads(dump)
+
+    assert_allclose(inc_ridge.coef_, inc_ridge_loaded.coef_, atol=1e-6)
+    assert_allclose(inc_ridge.intercept_, inc_ridge_loaded.intercept_, atol=1e-6)

sklearnex/linear_model/tests/test_linear.py

@@ -0,0 +1,142 @@
+# ===============================================================================
+# 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 numpy as np
+import pytest
+from numpy.testing import assert_allclose
+from scipy.linalg import lstsq
+from sklearn.datasets import make_regression
+
+from daal4py.sklearn._utils import daal_check_version
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex.tests.utils import _IS_INTEL
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.parametrize("macro_block", [None, 1024])
+@pytest.mark.parametrize("overdetermined", [False, True])
+@pytest.mark.parametrize("multi_output", [False, True])
+def test_sklearnex_import_linear(
+    dataframe, queue, dtype, macro_block, overdetermined, multi_output
+):
+    if (not overdetermined or multi_output) and not daal_check_version((2025, "P", 1)):
+        pytest.skip("Functionality introduced in later versions")
+    if (
+        not overdetermined
+        and queue
+        and queue.sycl_device.is_gpu
+        and not daal_check_version((2025, "P", 200))
+    ):
+        pytest.skip("Functionality introduced in later versions")
+
+    from sklearnex.linear_model import LinearRegression
+
+    rng = np.random.default_rng(seed=123)
+    X = rng.standard_normal(size=(10, 20) if not overdetermined else (20, 5))
+    y = rng.standard_normal(size=(X.shape[0], 3) if multi_output else X.shape[0])
+
+    Xi = np.c_[X, np.ones((X.shape[0], 1))]
+    expected_coefs = lstsq(Xi, y)[0]
+    expected_intercept = expected_coefs[-1]
+    expected_coefs = expected_coefs[: X.shape[1]]
+    if multi_output:
+        expected_coefs = expected_coefs.T
+
+    linreg = LinearRegression()
+    if daal_check_version((2024, "P", 0)) and macro_block is not None:
+        hparams = LinearRegression.get_hyperparameters("fit")
+        hparams.cpu_macro_block = macro_block
+        hparams.gpu_macro_block = macro_block
+
+    X = X.astype(dtype=dtype)
+    y = y.astype(dtype=dtype)
+    y_list = y.tolist()
+    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    linreg.fit(X, y)
+
+    assert hasattr(linreg, "_onedal_estimator")
+    assert "sklearnex" in linreg.__module__
+
+    rtol = 1e-3 if dtype == np.float32 else 1e-5
+    assert_allclose(_as_numpy(linreg.coef_), expected_coefs, rtol=rtol)
+    assert_allclose(_as_numpy(linreg.intercept_), expected_intercept, rtol=rtol)
+
+    # check that it also works with lists
+    if isinstance(X, np.ndarray):
+        linreg_list = LinearRegression().fit(X, y_list)
+        assert_allclose(linreg_list.coef_, linreg.coef_)
+        assert_allclose(linreg_list.intercept_, linreg.intercept_)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_sklearnex_import_lasso(dataframe, queue):
+    from sklearnex.linear_model import Lasso
+
+    X = [[0, 0], [1, 1], [2, 2]]
+    y = [0, 1, 2]
+    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    lasso = Lasso(alpha=0.1).fit(X, y)
+    assert "daal4py" in lasso.__module__
+    assert_allclose(lasso.intercept_, 0.15)
+    assert_allclose(lasso.coef_, [0.85, 0.0])
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_sklearnex_import_elastic(dataframe, queue):
+    from sklearnex.linear_model import ElasticNet
+
+    X, y = make_regression(n_features=2, random_state=0)
+    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    elasticnet = ElasticNet(random_state=0).fit(X, y)
+    assert "daal4py" in elasticnet.__module__
+    assert_allclose(elasticnet.intercept_, 1.451, atol=1e-3)
+    assert_allclose(elasticnet.coef_, [18.838, 64.559], atol=1e-3)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_sklearnex_reconstruct_model(dataframe, queue, dtype):
+    from sklearnex.linear_model import LinearRegression
+
+    seed = 42
+    num_samples = 3500
+    num_features, num_targets = 14, 9
+
+    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)
+    gtr = X @ coef + intercept[np.newaxis, :]
+
+    X = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+
+    linreg = LinearRegression(fit_intercept=True)
+    linreg.coef_ = coef.T
+    linreg.intercept_ = intercept
+
+    y_pred = linreg.predict(X)
+
+    tol = 1e-5 if _as_numpy(y_pred).dtype == np.float32 else 1e-7
+    assert_allclose(gtr, _as_numpy(y_pred), rtol=tol)

sklearnex/linear_model/tests/test_logreg.py

@@ -0,0 +1,134 @@
+# ===============================================================================
+# 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 numpy as np
+import pytest
+from numpy.testing import assert_allclose, assert_array_equal
+from scipy.sparse import csr_matrix
+from sklearn.datasets import load_breast_cancer, load_iris, make_classification
+from sklearn.metrics import accuracy_score
+from sklearn.model_selection import train_test_split
+
+from daal4py.sklearn._utils import daal_check_version
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+    get_queues,
+)
+from sklearnex import config_context
+
+
+def prepare_input(X, y, dataframe, queue):
+    X_train, X_test, y_train, y_test = train_test_split(
+        X, y, train_size=0.8, random_state=42
+    )
+    X_train = _convert_to_dataframe(X_train, sycl_queue=queue, target_df=dataframe)
+    y_train = _convert_to_dataframe(y_train, sycl_queue=queue, target_df=dataframe)
+    X_test = _convert_to_dataframe(X_test, sycl_queue=queue, target_df=dataframe)
+    return X_train, X_test, y_train, y_test
+
+
+@pytest.mark.parametrize(
+    "dataframe,queue", get_dataframes_and_queues(device_filter_="cpu")
+)
+def test_sklearnex_multiclass_classification(dataframe, queue):
+    from sklearnex.linear_model import LogisticRegression
+
+    X, y = load_iris(return_X_y=True)
+    X_train, X_test, y_train, y_test = prepare_input(X, y, dataframe, queue)
+
+    logreg = LogisticRegression(fit_intercept=True, solver="lbfgs", max_iter=200).fit(
+        X_train, y_train
+    )
+
+    if daal_check_version((2024, "P", 1)):
+        assert "sklearnex" in logreg.__module__
+    else:
+        assert "daal4py" in logreg.__module__
+
+    y_pred = _as_numpy(logreg.predict(X_test))
+    assert accuracy_score(y_test, y_pred) > 0.99
+
+
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(),
+)
+def test_sklearnex_binary_classification(dataframe, queue):
+    from sklearnex.linear_model import LogisticRegression
+
+    X, y = load_breast_cancer(return_X_y=True)
+    X_train, X_test, y_train, y_test = prepare_input(X, y, dataframe, queue)
+
+    logreg = LogisticRegression(fit_intercept=True, solver="newton-cg", max_iter=100).fit(
+        X_train, y_train
+    )
+
+    if daal_check_version((2024, "P", 1)):
+        assert "sklearnex" in logreg.__module__
+    else:
+        assert "daal4py" in logreg.__module__
+    if (
+        dataframe != "numpy"
+        and queue is not None
+        and queue.sycl_device.is_gpu
+        and daal_check_version((2024, "P", 1))
+    ):
+        # fit was done on gpu
+        assert hasattr(logreg, "_onedal_estimator")
+
+    y_pred = _as_numpy(logreg.predict(X_test))
+    assert accuracy_score(y_test, y_pred) > 0.95
+
+
+if daal_check_version((2024, "P", 700)):
+
+    @pytest.mark.parametrize("queue", get_queues("gpu"))
+    @pytest.mark.parametrize("dtype", [np.float32, np.float64])
+    @pytest.mark.parametrize(
+        "dims", [(3007, 17, 0.05), (50000, 100, 0.01), (512, 10, 0.5)]
+    )
+    def test_csr(queue, dtype, dims):
+        from sklearnex.linear_model import LogisticRegression
+
+        n, p, density = dims
+
+        # Create sparse dataset for classification
+        X, y = make_classification(n, p, random_state=42)
+        X = X.astype(dtype)
+        y = y.astype(dtype)
+        np.random.seed(2007 + n + p)
+        mask = np.random.binomial(1, density, (n, p))
+        X = X * mask
+        X_sp = csr_matrix(X)
+
+        model = LogisticRegression(fit_intercept=True, solver="newton-cg")
+        model_sp = LogisticRegression(fit_intercept=True, solver="newton-cg")
+
+        with config_context(target_offload="gpu:0"):
+            model.fit(X, y)
+            pred = model.predict(X)
+            prob = model.predict_proba(X)
+            model_sp.fit(X_sp, y)
+            pred_sp = model_sp.predict(X_sp)
+            prob_sp = model_sp.predict_proba(X_sp)
+
+        rtol = 2e-4
+        assert_allclose(pred, pred_sp, rtol=rtol)
+        assert_allclose(prob, prob_sp, rtol=rtol)
+        assert_allclose(model.coef_, model_sp.coef_, rtol=rtol)
+        assert_allclose(model.intercept_, model_sp.intercept_, rtol=rtol)