scikit-learn-intelex 2025.1.0__py39-none-manylinux_2_28_x86_64.whl

sklearnex/spmd/linear_model/tests/test_linear_regression_spmd.py

@@ -0,0 +1,145 @@
+# ==============================================================================
+# 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 (
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex.tests.utils.spmd import (
+    _generate_regression_data,
+    _get_local_tensor,
+    _mpi_libs_and_gpu_available,
+    _spmd_assert_allclose,
+)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.mpi
+def test_linear_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.linear_model import LinearRegression as LinearRegression_Batch
+    from sklearnex.spmd.linear_model import LinearRegression as LinearRegression_SPMD
+
+    # Create gold data and convert to dataframe
+    X_train = np.array(
+        [
+            [0.0, 0.0],
+            [0.0, 1.0],
+            [1.0, 0.0],
+            [0.0, 2.0],
+            [2.0, 0.0],
+            [1.0, 1.0],
+            [0.0, -1.0],
+            [-1.0, 0.0],
+            [-1.0, -1.0],
+        ]
+    )
+    y_train = np.array([3.0, 5.0, 4.0, 7.0, 5.0, 6.0, 1.0, 2.0, 0.0])
+    X_test = np.array(
+        [
+            [1.0, -1.0],
+            [-1.0, 1.0],
+            [0.0, 1.0],
+            [10.0, -10.0],
+        ]
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # ensure trained model of batch algo matches spmd
+    spmd_model = LinearRegression_SPMD().fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = LinearRegression_Batch().fit(X_train, y_train)
+
+    assert_allclose(spmd_model.coef_, batch_model.coef_)
+    assert_allclose(spmd_model.intercept_, batch_model.intercept_)
+
+    # ensure predictions of batch algo match spmd
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _spmd_assert_allclose(spmd_result, batch_result)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize("n_samples", [100, 10000])
+@pytest.mark.parametrize("n_features", [10, 100])
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_linear_spmd_synthetic(n_samples, n_features, dataframe, queue, dtype):
+    # Import spmd and batch algo
+    from sklearnex.linear_model import LinearRegression as LinearRegression_Batch
+    from sklearnex.spmd.linear_model import LinearRegression as LinearRegression_SPMD
+
+    # Generate data and convert to dataframe
+    X_train, X_test, y_train, _ = _generate_regression_data(
+        n_samples, n_features, dtype=dtype
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # TODO: support linear regression on wide datasets and remove this skip
+    if local_dpt_X_train.shape[0] < n_features:
+        pytest.skip(
+            "SPMD Linear Regression does not support cases where n_rows_rank < n_features"
+        )
+
+    # ensure trained model of batch algo matches spmd
+    spmd_model = LinearRegression_SPMD().fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = LinearRegression_Batch().fit(X_train, y_train)
+
+    tol = 1e-3 if dtype == np.float32 else 1e-7
+    assert_allclose(spmd_model.coef_, batch_model.coef_, rtol=tol, atol=tol)
+    assert_allclose(spmd_model.intercept_, batch_model.intercept_, rtol=tol, atol=tol)
+
+    # ensure predictions of batch algo match spmd
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _spmd_assert_allclose(spmd_result, batch_result, rtol=tol, atol=tol)

sklearnex/spmd/linear_model/tests/test_logistic_regression_spmd.py

@@ -0,0 +1,162 @@
+# ==============================================================================
+# 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.tests.utils.spmd import (
+    _generate_classification_data,
+    _get_local_tensor,
+    _mpi_libs_and_gpu_available,
+    _spmd_assert_allclose,
+)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.mpi
+def test_logistic_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.linear_model import LogisticRegression as LogisticRegression_Batch
+    from sklearnex.spmd.linear_model import LogisticRegression as LogisticRegression_SPMD
+
+    # Create gold data and convert to dataframe
+    X_train = np.array(
+        [
+            [0.0, 0.0],
+            [0.0, 1.0],
+            [1.0, 0.0],
+            [0.0, 2.0],
+            [2.0, 0.0],
+            [1.0, 1.0],
+            [0.0, -1.0],
+            [-1.0, 0.0],
+            [-1.0, -1.0],
+        ]
+    )
+    y_train = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0])
+    X_test = np.array(
+        [
+            [1.0, -1.0],
+            [-1.0, 1.0],
+            [0.0, 1.0],
+            [10.0, -10.0],
+        ]
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+    dpt_X_train = _convert_to_dataframe(X_train, sycl_queue=queue, target_df=dataframe)
+    dpt_y_train = _convert_to_dataframe(y_train, sycl_queue=queue, target_df=dataframe)
+    dpt_X_test = _convert_to_dataframe(X_test, sycl_queue=queue, target_df=dataframe)
+
+    # Ensure trained model of batch algo matches spmd
+    spmd_model = LogisticRegression_SPMD(random_state=0, solver="newton-cg").fit(
+        local_dpt_X_train, local_dpt_y_train
+    )
+    batch_model = LogisticRegression_Batch(random_state=0, solver="newton-cg").fit(
+        dpt_X_train, dpt_y_train
+    )
+
+    assert_allclose(spmd_model.coef_, batch_model.coef_, rtol=1e-2)
+    assert_allclose(spmd_model.intercept_, batch_model.intercept_, rtol=1e-2)
+
+    # Ensure predictions of batch algo match spmd
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(dpt_X_test)
+
+    _spmd_assert_allclose(spmd_result, _as_numpy(batch_result))
+
+
+# parametrize max_iter, C, tol
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize("n_samples", [100, 10000])
+@pytest.mark.parametrize("n_features", [10, 100])
+@pytest.mark.parametrize("C", [0.5, 1.0, 2.0])
+@pytest.mark.parametrize("tol", [1e-2, 1e-4])
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_logistic_spmd_synthetic(n_samples, n_features, C, tol, dataframe, queue, dtype):
+    # TODO: Resolve numerical issues when n_rows_rank < n_cols
+    if n_samples <= n_features:
+        pytest.skip("Numerical issues when rank rows < columns")
+
+    # Import spmd and batch algo
+    from sklearnex.linear_model import LogisticRegression as LogisticRegression_Batch
+    from sklearnex.spmd.linear_model import LogisticRegression as LogisticRegression_SPMD
+
+    # Generate data and convert to dataframe
+    X_train, X_test, y_train, _ = _generate_classification_data(
+        n_samples, n_features, dtype=dtype
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+    dpt_X_train = _convert_to_dataframe(X_train, sycl_queue=queue, target_df=dataframe)
+    dpt_y_train = _convert_to_dataframe(y_train, sycl_queue=queue, target_df=dataframe)
+    dpt_X_test = _convert_to_dataframe(X_test, sycl_queue=queue, target_df=dataframe)
+
+    # Ensure trained model of batch algo matches spmd
+    spmd_model = LogisticRegression_SPMD(
+        random_state=0, solver="newton-cg", C=C, tol=tol
+    ).fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = LogisticRegression_Batch(
+        random_state=0, solver="newton-cg", C=C, tol=tol
+    ).fit(dpt_X_train, dpt_y_train)
+
+    # TODO: Logistic Regression coefficients do not align
+    tol = 1e-2
+    assert_allclose(spmd_model.coef_, batch_model.coef_, rtol=tol, atol=tol)
+    assert_allclose(spmd_model.intercept_, batch_model.intercept_, rtol=tol, atol=tol)
+
+    # Ensure predictions of batch algo match spmd
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(dpt_X_test)
+
+    _spmd_assert_allclose(spmd_result, _as_numpy(batch_result))

sklearnex/spmd/neighbors/__init__.py

@@ -0,0 +1,19 @@
+# ==============================================================================
+# Copyright 2023 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 .neighbors import KNeighborsClassifier, KNeighborsRegressor, NearestNeighbors
+
+__all__ = ["KNeighborsClassifier", "KNeighborsRegressor", "NearestNeighbors"]

sklearnex/spmd/neighbors/neighbors.py

@@ -0,0 +1,25 @@
+# ==============================================================================
+# Copyright 2023 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 onedal.spmd.neighbors import (
+    KNeighborsClassifier,
+    KNeighborsRegressor,
+    NearestNeighbors,
+)
+
+# TODO:
+# Currently it uses `onedal` module interface.
+# Add sklearnex dispatching.

sklearnex/spmd/neighbors/tests/test_neighbors_spmd.py

@@ -0,0 +1,288 @@
+# ==============================================================================
+# 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 (
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex.tests.utils.spmd import (
+    _assert_unordered_allclose,
+    _generate_classification_data,
+    _generate_regression_data,
+    _get_local_tensor,
+    _mpi_libs_and_gpu_available,
+    _spmd_assert_allclose,
+)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.mpi
+def test_knncls_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.neighbors import KNeighborsClassifier as KNeighborsClassifier_Batch
+    from sklearnex.spmd.neighbors import KNeighborsClassifier as KNeighborsClassifier_SPMD
+
+    # Create gold data and convert to dataframe
+    X_train = np.array(
+        [
+            [0.0, 0.0],
+            [0.0, 1.0],
+            [1.0, 0.0],
+            [0.0, 2.0],
+            [2.0, 0.0],
+            [0.9, 1.0],
+            [0.0, -1.0],
+            [-1.0, 0.0],
+            [-1.0, -1.0],
+        ]
+    )
+    # TODO: handle situations where not all classes are present on all ranks?
+    y_train = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0])
+    X_test = np.array(
+        [
+            [1.0, -0.5],
+            [-5.0, 1.0],
+            [0.0, 1.0],
+            [10.0, -10.0],
+        ]
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure predictions of batch algo match spmd
+    spmd_model = KNeighborsClassifier_SPMD(n_neighbors=1, algorithm="brute").fit(
+        local_dpt_X_train, local_dpt_y_train
+    )
+    batch_model = KNeighborsClassifier_Batch(n_neighbors=1, algorithm="brute").fit(
+        X_train, y_train
+    )
+    spmd_dists, spmd_indcs = spmd_model.kneighbors(local_dpt_X_test)
+    batch_dists, batch_indcs = batch_model.kneighbors(X_test)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _assert_unordered_allclose(spmd_indcs, batch_indcs, localize=True)
+    _assert_unordered_allclose(spmd_dists, batch_dists, localize=True)
+    _spmd_assert_allclose(spmd_result, batch_result)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize("n_samples", [200, 10000])
+@pytest.mark.parametrize("n_features_and_classes", [(5, 2), (25, 2), (25, 10)])
+@pytest.mark.parametrize("n_neighbors", [1, 5, 20])
+@pytest.mark.parametrize("weights", ["uniform", "distance"])
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_knncls_spmd_synthetic(
+    n_samples,
+    n_features_and_classes,
+    n_neighbors,
+    weights,
+    dataframe,
+    queue,
+    dtype,
+    metric="euclidean",
+):
+    n_features, n_classes = n_features_and_classes
+    # Import spmd and batch algo
+    from sklearnex.neighbors import KNeighborsClassifier as KNeighborsClassifier_Batch
+    from sklearnex.spmd.neighbors import KNeighborsClassifier as KNeighborsClassifier_SPMD
+
+    # Generate data and convert to dataframe
+    X_train, X_test, y_train, _ = _generate_classification_data(
+        n_samples, n_features, n_classes, dtype=dtype
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure predictions of batch algo match spmd
+    spmd_model = KNeighborsClassifier_SPMD(
+        n_neighbors=n_neighbors, weights=weights, metric=metric, algorithm="brute"
+    ).fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = KNeighborsClassifier_Batch(
+        n_neighbors=n_neighbors, weights=weights, metric=metric, algorithm="brute"
+    ).fit(X_train, y_train)
+    spmd_dists, spmd_indcs = spmd_model.kneighbors(local_dpt_X_test)
+    batch_dists, batch_indcs = batch_model.kneighbors(X_test)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    tol = 1e-4
+    if dtype == np.float64:
+        _assert_unordered_allclose(spmd_indcs, batch_indcs, localize=True)
+        _assert_unordered_allclose(
+            spmd_dists, batch_dists, localize=True, rtol=tol, atol=tol
+        )
+    _spmd_assert_allclose(spmd_result, batch_result)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.mpi
+def test_knnreg_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.neighbors import KNeighborsRegressor as KNeighborsRegressor_Batch
+    from sklearnex.spmd.neighbors import KNeighborsRegressor as KNeighborsRegressor_SPMD
+
+    # Create gold data and convert to dataframe
+    X_train = np.array(
+        [
+            [0.0, 0.0],
+            [0.0, 1.0],
+            [1.0, 0.0],
+            [0.0, 2.0],
+            [2.0, 0.0],
+            [1.0, 1.0],
+            [0.0, -1.0],
+            [-1.0, 0.0],
+            [-1.0, -1.0],
+        ]
+    )
+    y_train = np.array([3.0, 5.0, 4.0, 7.0, 5.0, 6.0, 1.0, 2.0, 0.0])
+    X_test = np.array(
+        [
+            [1.0, -0.5],
+            [-5.0, 1.0],
+            [0.0, 1.0],
+            [10.0, -10.0],
+        ]
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure predictions of batch algo match spmd
+    spmd_model = KNeighborsRegressor_SPMD(n_neighbors=1, algorithm="brute").fit(
+        local_dpt_X_train, local_dpt_y_train
+    )
+    batch_model = KNeighborsRegressor_Batch(n_neighbors=1, algorithm="brute").fit(
+        X_train, y_train
+    )
+    spmd_dists, spmd_indcs = spmd_model.kneighbors(local_dpt_X_test)
+    batch_dists, batch_indcs = batch_model.kneighbors(X_test)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _assert_unordered_allclose(spmd_indcs, batch_indcs, localize=True)
+    _assert_unordered_allclose(spmd_dists, batch_dists, localize=True)
+    _spmd_assert_allclose(spmd_result, batch_result)
+
+
+@pytest.mark.skipif(
+    not _mpi_libs_and_gpu_available,
+    reason="GPU device and MPI libs required for test",
+)
+@pytest.mark.parametrize("n_samples", [200, 10000])
+@pytest.mark.parametrize("n_features", [5, 25])
+@pytest.mark.parametrize("n_neighbors", [1, 5, 20])
+@pytest.mark.parametrize("weights", ["uniform", "distance"])
+@pytest.mark.parametrize(
+    "metric", ["euclidean", "manhattan", "minkowski", "chebyshev", "cosine"]
+)
+@pytest.mark.parametrize(
+    "dataframe,queue",
+    get_dataframes_and_queues(dataframe_filter_="dpnp,dpctl", device_filter_="gpu"),
+)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_knnreg_spmd_synthetic(
+    n_samples, n_features, n_neighbors, weights, metric, dataframe, queue, dtype
+):
+    # Import spmd and batch algo
+    from sklearnex.neighbors import KNeighborsRegressor as KNeighborsRegressor_Batch
+    from sklearnex.spmd.neighbors import KNeighborsRegressor as KNeighborsRegressor_SPMD
+
+    # Generate data and convert to dataframe
+    X_train, X_test, y_train, _ = _generate_regression_data(
+        n_samples, n_features, dtype=dtype
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_y_train = _convert_to_dataframe(
+        _get_local_tensor(y_train), sycl_queue=queue, target_df=dataframe
+    )
+    local_dpt_X_test = _convert_to_dataframe(
+        _get_local_tensor(X_test), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure predictions of batch algo match spmd
+    spmd_model = KNeighborsRegressor_SPMD(
+        n_neighbors=n_neighbors, weights=weights, metric=metric, algorithm="brute"
+    ).fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = KNeighborsRegressor_Batch(
+        n_neighbors=n_neighbors, weights=weights, metric=metric, algorithm="brute"
+    ).fit(X_train, y_train)
+    spmd_dists, spmd_indcs = spmd_model.kneighbors(local_dpt_X_test)
+    batch_dists, batch_indcs = batch_model.kneighbors(X_test)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    tol = 0.005 if dtype == np.float32 else 1e-4
+    if dtype == np.float64:
+        _assert_unordered_allclose(spmd_indcs, batch_indcs, localize=True)
+        _assert_unordered_allclose(
+            spmd_dists, batch_dists, localize=True, rtol=tol, atol=tol
+        )
+    _spmd_assert_allclose(spmd_result, batch_result, rtol=tol, atol=tol)

sklearnex/svm/__init__.py

@@ -0,0 +1,29 @@
+# ==============================================================================
+# 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 .._utils import get_sklearnex_version
+
+if get_sklearnex_version((2021, "P", 300)):
+    from .nusvc import NuSVC
+    from .nusvr import NuSVR
+    from .svc import SVC
+    from .svr import SVR
+
+    __all__ = ["SVR", "SVC", "NuSVC", "NuSVR"]
+else:
+    from daal4py.sklearn.svm import SVC
+
+    __all__ = ["SVC"]