scikit-learn-intelex 2024.4.0__py312-none-win_amd64.whl → 2025.10.0__py312-none-win_amd64.whl

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/linear_model/tests/test_ridge.py

@@ -0,0 +1,256 @@
+# ===============================================================================
+# 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 sklearn.exceptions import NotFittedError
+
+from daal4py.sklearn._utils import daal_check_version
+from daal4py.sklearn.linear_model.tests.test_ridge import (
+    _test_multivariate_ridge_alpha_shape,
+    _test_multivariate_ridge_coefficients,
+)
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+
+
+def _compute_ridge_coefficients(X, y, alpha, fit_intercept):
+    """
+    Computes Ridge regression coefficients for single or multiple target variables.
+
+    Parameters:
+    X : array-like of shape (n_samples, n_features)
+        Training data.
+    y : array-like of shape (n_samples,) or (n_samples, n_targets)
+        Target values.
+    alpha : float
+        Regularization strength.
+    fit_intercept : bool
+        Whether to calculate the intercept for this model.
+
+    Returns:
+    coeffs : ndarray of shape (n_features, n_targets)
+        Estimated coefficients for each target variable.
+    intercept : ndarray of shape (n_targets,)
+        Intercept terms for each target variable.
+    """
+    n_samples, n_features = X.shape
+    X_copy = X.copy()
+    y = np.asarray(y)
+
+    # making single-target y also multi-dim array for consistency
+    if y.ndim == 1:
+        y = y[:, np.newaxis]
+    n_targets = y.shape[1]
+
+    if fit_intercept:
+        # adding column of ones to X for the intercept term
+        X_copy = np.hstack([np.ones((n_samples, 1)), X])
+        identity_matrix = np.diag([0] + [1] * n_features)
+    else:
+        identity_matrix = np.eye(n_features)
+
+    # normal equation: (X^T * X + alpha * I) * w = X^T * y
+    A = X_copy.T @ X_copy + alpha * identity_matrix
+    b = X_copy.T @ y
+    coeffs = np.linalg.solve(A, b)
+
+    if fit_intercept:
+        intercept = coeffs[0, :]  # Shape (n_targets,)
+        coeffs = coeffs[1:, :]  # Shape (n_features, n_targets)
+    else:
+        intercept = np.zeros((n_targets,))  # Shape (n_targets,)
+
+    # in case of single target, flattening both coefficients and intercept
+    if n_targets == 1:
+        coeffs = coeffs.flatten()
+        intercept = intercept[0]
+    else:
+        coeffs = coeffs.T
+
+    return coeffs, intercept
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_sklearnex_import_ridge(dataframe, queue):
+    from sklearnex.linear_model import Ridge
+
+    X = np.array([[1, 1], [1, 2], [2, 2], [2, 3]])
+    y = np.dot(X, np.array([1, 2])) + 3
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    ridge_reg = Ridge(alpha=0.5).fit(X_c, y_c)
+
+    if daal_check_version((2024, "P", 600)):
+        assert (
+            "sklearnex" in ridge_reg.__module__ and "preview" not in ridge_reg.__module__
+        )
+    else:
+        assert "daal4py" in ridge_reg.__module__
+
+    assert_allclose(ridge_reg.intercept_, 3.86, rtol=1e-2)
+    assert_allclose(ridge_reg.coef_, [0.91, 1.64], rtol=1e-2)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("sample_size", [100, 1000])
+@pytest.mark.parametrize("feature_size", [10, 50])
+@pytest.mark.parametrize("alpha", [0.1, 0.5, 1.0])
+@pytest.mark.parametrize("fit_intercept", [True, False])
+def test_ridge_coefficients(
+    dataframe, queue, sample_size, feature_size, alpha, fit_intercept
+):
+    from sklearnex.linear_model import Ridge
+
+    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)
+    ridge_reg = Ridge(fit_intercept=fit_intercept, alpha=alpha).fit(X_c, y_c)
+
+    coefficients_manual, intercept_manual = _compute_ridge_coefficients(
+        X, y, alpha, fit_intercept=fit_intercept
+    )
+
+    assert_allclose(ridge_reg.coef_, coefficients_manual, rtol=1e-6, atol=1e-6)
+    assert_allclose(ridge_reg.intercept_, intercept_manual, rtol=1e-6, atol=1e-6)
+
+
+@pytest.mark.skipif(
+    not daal_check_version((2024, "P", 600)), reason="requires onedal 2024.6.0 or higher"
+)
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_ridge_score_before_fit(dataframe, queue):
+    from sklearnex.linear_model import Ridge
+
+    sample_count, feature_count = 10, 5
+
+    model = Ridge(fit_intercept=True, alpha=0.5)
+
+    X, y = np.random.rand(sample_count, feature_count), np.random.rand(sample_count)
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+
+    with pytest.raises(NotFittedError):
+        model.score(X_c, y_c)
+
+
+@pytest.mark.skipif(
+    not daal_check_version((2024, "P", 600)), reason="requires onedal 2024.6.0 or higher"
+)
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_ridge_predict_before_fit(dataframe, queue):
+    from sklearnex.linear_model import Ridge
+
+    sample_count, feature_count = 10, 5
+
+    model = Ridge(fit_intercept=True, alpha=0.5)
+
+    X = np.random.rand(sample_count, feature_count)
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+
+    with pytest.raises(NotFittedError):
+        model.predict(X_c)
+
+
+@pytest.mark.skip(reason="Forced to use original sklearn for now.")
+def test_sklearnex_multivariate_ridge_coefs():
+    from sklearnex.linear_model import Ridge
+
+    _test_multivariate_ridge_coefficients(Ridge, random_state=0)
+
+
+@pytest.mark.skip(reason="Forced to use original sklearn for now.")
+def test_sklearnex_multivariate_ridge_alpha_shape():
+    from sklearnex.linear_model import Ridge
+
+    _test_multivariate_ridge_alpha_shape(Ridge, random_state=0)
+
+
+@pytest.mark.skipif(
+    not daal_check_version((2025, "P", 100)), reason="requires onedal 2025.1.0 or higher"
+)
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("overdetermined", [True, False])
+@pytest.mark.parametrize("alpha", [0.00001, 0.1, 1.0])
+@pytest.mark.parametrize("fit_intercept", [True, False])
+def test_ridge_overdetermined_system(
+    dataframe, queue, overdetermined, alpha, fit_intercept
+):
+    from sklearnex.linear_model import Ridge
+
+    if overdetermined:
+        X = np.random.rand(100, 10)
+        y = np.random.rand(100)
+    else:
+        X = np.random.rand(10, 100)
+        y = np.random.rand(10)
+
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    ridge_reg = Ridge(alpha=alpha, fit_intercept=fit_intercept).fit(X_c, y_c)
+
+    coefficients_manual, intercept_manual = _compute_ridge_coefficients(
+        X, y, alpha, fit_intercept=fit_intercept
+    )
+
+    assert_allclose(ridge_reg.coef_, coefficients_manual, rtol=1e-6, atol=1e-6)
+    assert_allclose(ridge_reg.intercept_, intercept_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("alpha", [0.1, 0.5, 1.0])
+def test_multivariate_ridge_scalar_alpha(dataframe, queue, fit_intercept, alpha):
+    from sklearn.datasets import make_regression
+
+    from sklearnex.linear_model import Ridge
+
+    X, y = make_regression(n_samples=10, n_features=3, n_targets=3, random_state=0)
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    ridge = Ridge(alpha=alpha, fit_intercept=fit_intercept)
+    ridge.fit(X_c, y_c)
+
+    coef_manual, intercept = _compute_ridge_coefficients(X, y, alpha, fit_intercept)
+    assert_allclose(ridge.coef_, coef_manual, rtol=1e-6, atol=1e-6)
+    assert_allclose(ridge.intercept_, intercept, rtol=1e-6, atol=1e-6)
+
+    predictions = _as_numpy(ridge.predict(X_c))
+    predictions_manual = X @ coef_manual.T + intercept
+    assert_allclose(predictions, predictions_manual, rtol=1e-6, atol=1e-6)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+def test_underdetermined_positive_alpha_ridge(dataframe, queue):
+    from sklearn.datasets import make_regression
+
+    from sklearnex.linear_model import Ridge
+
+    X, y = make_regression(n_samples=5, n_features=6, n_targets=1, random_state=0)
+    alpha = 1.0
+    X_c = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    y_c = _convert_to_dataframe(y, sycl_queue=queue, target_df=dataframe)
+    ridge = Ridge(alpha=alpha, fit_intercept=True).fit(X_c, y_c)
+
+    coef_manual, intercept = _compute_ridge_coefficients(X, y, alpha, fit_intercept=True)
+
+    assert_allclose(ridge.coef_, coef_manual, rtol=1e-6, atol=1e-6)
+    assert_allclose(ridge.intercept_, intercept, rtol=1e-6, atol=1e-6)

{scikit_learn_intelex-2024.4.0.data → scikit_learn_intelex-2025.10.0.data}/data/Lib/site-packages/sklearnex/manifold/t_sne.py

@@ -14,4 +14,15 @@
 # limitations under the License.
 # ===============================================================================
 
+from daal4py.sklearn._utils import sklearn_check_version
 from daal4py.sklearn.manifold import TSNE
+from onedal._device_offload import support_input_format
+
+from ..base import oneDALEstimator
+
+TSNE.fit = support_input_format(TSNE.fit)
+TSNE.fit_transform = support_input_format(TSNE.fit_transform)
+
+if sklearn_check_version("1.4"):
+    TSNE._doc_link_module = "daal4py"
+    TSNE._doc_link_url_param_generator = oneDALEstimator._doc_link_url_param_generator

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/manifold/tests/test_tsne.py

@@ -0,0 +1,256 @@
+# ===============================================================================
+# 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 sklearn.metrics.pairwise import pairwise_distances
+
+from daal4py.sklearn._utils import sklearn_check_version
+
+# Note: n_components must be 2 for now
+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):
+    """Test TSNE compatibility with different backends and queues, and validate sklearnex module."""
+    from sklearnex.manifold import TSNE
+
+    X = np.array([[0, 0, 0], [0, 1, 1], [1, 0, 1], [1, 1, 1]])
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    tsne = TSNE(n_components=2, perplexity=2.0, random_state=42, init="pca").fit(X_df)
+    embedding = tsne.fit_transform(X_df)
+    embedding = _as_numpy(embedding)
+    assert "daal4py" in tsne.__module__
+    assert tsne.n_components == 2
+    assert tsne.perplexity == 2.0
+    assert tsne.random_state == 42
+    assert tsne.init == "pca"
+
+
+@pytest.mark.parametrize(
+    "X_generator,n_components,perplexity,expected_shape,should_raise",
+    [
+        pytest.param(
+            lambda rng: np.array([[0, 0, 0], [0, 1, 1], [1, 0, 1], [1, 1, 1]]),
+            2,
+            2.0,
+            (4, 2),
+            False,
+            id="Basic functionality",
+        ),
+        pytest.param(
+            lambda rng: rng.random((100, 10)),
+            2,
+            30.0,
+            (100, 2),
+            False,
+            id="Random data",
+        ),
+        pytest.param(
+            lambda rng: np.array([[0, 0], [1, 1], [2, 2]]),
+            2,
+            2.0,
+            (3, 2),
+            False,
+            id="Valid minimal data",
+        ),
+        pytest.param(
+            lambda rng: np.empty((0, 10)),
+            2,
+            5.0,
+            None,
+            True,
+            id="Empty data",
+        ),
+        pytest.param(
+            lambda rng: np.array([[0, 0], [1, np.nan], [2, np.inf]]),
+            2,
+            5.0,
+            None,
+            True,
+            id="Data with NaN/Inf",
+        ),
+        pytest.param(
+            lambda rng: rng.random((50, 500)) * (rng.random((50, 500)) > 0.99),
+            2,
+            30.0,
+            (50, 2),
+            False,
+            id="Sparse-like high-dimensional data",
+        ),
+        pytest.param(
+            lambda rng: np.hstack(
+                [
+                    np.ones((50, 1)),  # First column is 1
+                    rng.random((50, 499)) * (rng.random((50, 499)) > 0.99),
+                ]
+            ),
+            2,
+            30.0,
+            (50, 2),
+            False,
+            id="Sparse-like data with constant column",
+        ),
+        pytest.param(
+            lambda rng: np.where(
+                np.arange(50 * 500).reshape(50, 500) % 10 == 0, 0, rng.random((50, 500))
+            ),
+            2,
+            30.0,
+            (50, 2),
+            False,
+            id="Sparse-like data with every tenth element zero",
+        ),
+        pytest.param(
+            lambda rng: rng.random((10, 5)),
+            2,
+            0.5,
+            (10, 2),
+            False,
+            id="Extremely low perplexity",
+        ),
+    ],
+)
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_tsne_functionality_and_edge_cases(
+    X_generator,
+    n_components,
+    perplexity,
+    expected_shape,
+    should_raise,
+    dataframe,
+    queue,
+    dtype,
+):
+    from sklearnex.manifold import TSNE
+
+    rng = np.random.default_rng(
+        seed=42
+    )  # Use generator to ensure independent dataset per test
+    X = X_generator(rng)
+    X = X.astype(dtype) if X.size > 0 else X
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+
+    if should_raise:
+        with pytest.raises(ValueError):
+            TSNE(n_components=n_components, perplexity=perplexity).fit_transform(X_df)
+    else:
+        tsne = TSNE(n_components=n_components, perplexity=perplexity, random_state=42)
+        embedding = tsne.fit_transform(X_df)
+        embedding = _as_numpy(embedding)
+        assert embedding.shape == expected_shape
+        assert np.all(np.isfinite(embedding))
+        assert np.any(embedding != 0)
+
+
+# Note: since sklearn1.2, the PCA initialization divides by standard deviations of components.
+# Since those will be zeros for constant data, it will end up producing NaNs, hence it's not tested.
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize(
+    "init", ["random"] + (["pca"] if not sklearn_check_version("1.2") else [])
+)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_tsne_constant_data(init, dataframe, queue, dtype):
+    from sklearnex.manifold import TSNE
+
+    X = np.ones((10, 10), dtype=dtype)
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    tsne = TSNE(n_components=2, init=init, perplexity=5, random_state=42)
+    embedding = tsne.fit_transform(X_df)
+    embedding = _as_numpy(embedding)
+    assert embedding.shape == (10, 2)
+    if init == "pca":
+        assert np.isclose(embedding[:, 0].std(), 0, atol=1e-6)  # Constant first dimension
+        assert np.allclose(embedding[:, 1], 0, atol=1e-6)  # Zero second dimension
+    elif init == "random":
+        assert np.all(np.isfinite(embedding))
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_tsne_reproducibility(dataframe, queue, dtype):
+    from sklearnex.manifold import TSNE
+
+    rng = np.random.default_rng(seed=42)
+    X = rng.random((50, 10)).astype(dtype)
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    tsne_1 = TSNE(n_components=2, random_state=42).fit_transform(X_df)
+    tsne_2 = TSNE(n_components=2, random_state=42).fit_transform(X_df)
+    # in case of dpctl.tensor.usm_ndarray convert to numpy array
+    tsne_1 = _as_numpy(tsne_1)
+    tsne_2 = _as_numpy(tsne_2)
+    assert_allclose(tsne_1, tsne_2, rtol=1e-5)
+
+
+@pytest.mark.parametrize("dataframe,queue", get_dataframes_and_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_tsne_complex_and_gpu_validation(dataframe, queue, dtype):
+    from sklearnex.manifold import TSNE
+
+    X = np.array(
+        [
+            [1, 1, 1, 1],
+            [1.1, 1.1, 1.1, 1.1],
+            [0.9, 0.9, 0.9, 0.9],
+            [2e9, 2e-9, -2e9, -2e-9],
+            [5e-5, 5e5, -5e-5, -5e5],
+            [9e-7, -9e7, 9e-7, -9e7],
+            [1, -1, 1, -1],
+            [-1e-9, 1e-9, -1e-9, 1e-9],
+            [42, 42, 42, 42],
+            [8, -8, 8e8, -8e-8],
+            [1e-3, 1e3, -1e3, -1e-3],
+            [0, 1e9, -1e-9, 1],
+            [0, 0, 1, -1],
+            [0, 0, 0, 0],
+            [-1e5, 0, 1e5, -1],
+            [1, 0, -1e8, 1e8],
+        ]
+    )
+    n_components = 2
+    perplexity = 3.0
+    expected_shape = (16, 2)
+
+    X = X.astype(dtype)
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+    tsne = TSNE(n_components=n_components, perplexity=perplexity, random_state=42)
+    embedding = tsne.fit_transform(X_df)
+
+    # Validate results
+    assert embedding.shape == expected_shape
+    embedding = _as_numpy(embedding)
+    assert np.all(np.isfinite(embedding))
+    assert np.any(embedding != 0)
+
+    # Ensure close points in original space remain close in embedding
+    group_a_indices = [0, 1, 2]  # Hardcoded index of similar points
+    group_b_indices = [3, 4, 5]  # Hardcoded index of dissimilar points from a
+    embedding_distances = pairwise_distances(
+        X, metric="euclidean"
+    )  # Get an array of distance where [i, j] is distance b/t i and j
+    # Check for distance b/t two points in group A < distance of this point and any point in group B
+    for i in group_a_indices:
+        for j in group_a_indices:
+            assert (
+                embedding_distances[i, j] < embedding_distances[i, group_b_indices].min()
+            ), f"Point {i} in Group A is closer to a point in Group B than to another point in Group A."

{scikit_learn_intelex-2024.4.0.data → scikit_learn_intelex-2025.10.0.data}/data/Lib/site-packages/sklearnex/metrics/pairwise.py

@@ -15,3 +15,6 @@
 # ===============================================================================
 
 from daal4py.sklearn.metrics import pairwise_distances
+from onedal._device_offload import support_input_format
+
+pairwise_distances = support_input_format(pairwise_distances)

{scikit_learn_intelex-2024.4.0.data → scikit_learn_intelex-2025.10.0.data}/data/Lib/site-packages/sklearnex/metrics/ranking.py

@@ -15,3 +15,6 @@
 # ===============================================================================
 
 from daal4py.sklearn.metrics import roc_auc_score
+from onedal._device_offload import support_input_format
+
+roc_auc_score = support_input_format(roc_auc_score)

{scikit_learn_intelex-2024.4.0.data → scikit_learn_intelex-2025.10.0.data}/data/Lib/site-packages/sklearnex/model_selection/split.py

@@ -15,3 +15,6 @@
 # ===============================================================================
 
 from daal4py.sklearn.model_selection import train_test_split
+from onedal._device_offload import support_input_format
+
+train_test_split = support_input_format(train_test_split)