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

sklearnex/spmd/decomposition/tests/test_pca_spmd.py

@@ -0,0 +1,128 @@
+# ==============================================================================
+# 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_statistic_data,
+    _get_local_tensor,
+    _mpi_libs_and_gpu_available,
+)
+
+
+@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_pca_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.decomposition import PCA as PCA_Batch
+    from sklearnex.spmd.decomposition import PCA as PCA_SPMD
+
+    # Create gold data and convert to dataframe
+    data = np.array(
+        [
+            [0.0, 0.0, 0.0],
+            [0.0, 1.0, 2.0],
+            [0.0, 2.0, 4.0],
+            [0.0, 3.0, 8.0],
+            [0.0, 4.0, 16.0],
+            [0.0, 5.0, 32.0],
+            [0.0, 6.0, 64.0],
+            [0.0, 7.0, 128.0],
+        ]
+    )
+
+    local_dpt_data = _convert_to_dataframe(
+        _get_local_tensor(data), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure results of batch algo match spmd
+    spmd_result = PCA_SPMD(n_components=2).fit(local_dpt_data)
+    batch_result = PCA_Batch(n_components=2).fit(data)
+
+    assert_allclose(spmd_result.mean_, batch_result.mean_)
+    assert_allclose(spmd_result.components_, batch_result.components_)
+    assert_allclose(spmd_result.singular_values_, batch_result.singular_values_)
+    assert_allclose(
+        spmd_result.noise_variance_,
+        batch_result.noise_variance_,
+        atol=1e-7,
+    )
+    assert_allclose(
+        spmd_result.explained_variance_ratio_, batch_result.explained_variance_ratio_
+    )
+
+
+@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("n_components", [0.5, 3, "mle", None])
+@pytest.mark.parametrize("whiten", [True, False])
+@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_pca_spmd_synthetic(
+    n_samples, n_features, n_components, whiten, dataframe, queue, dtype
+):
+    # TODO: Resolve issues with batch fallback and lack of support for n_rows_rank < n_cols
+    if n_components == "mle" or n_components == 3:
+        pytest.skip("Avoid error in case of batch fallback to sklearn")
+    if n_samples <= n_features:
+        pytest.skip("Avoid n_samples < n_features error from spmd data split")
+
+    # Import spmd and batch algo
+    from sklearnex.decomposition import PCA as PCA_Batch
+    from sklearnex.spmd.decomposition import PCA as PCA_SPMD
+
+    # Generate data and convert to dataframe
+    data = _generate_statistic_data(n_samples, n_features, dtype=dtype)
+
+    local_dpt_data = _convert_to_dataframe(
+        _get_local_tensor(data), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure results of batch algo match spmd
+    spmd_result = PCA_SPMD(n_components=n_components, whiten=whiten).fit(local_dpt_data)
+    batch_result = PCA_Batch(n_components=n_components, whiten=whiten).fit(data)
+
+    tol = 1e-3 if dtype == np.float32 else 1e-7
+    assert_allclose(spmd_result.mean_, batch_result.mean_, atol=tol)
+    assert_allclose(spmd_result.components_, batch_result.components_, atol=tol, rtol=tol)
+    assert_allclose(spmd_result.singular_values_, batch_result.singular_values_, atol=tol)
+    assert_allclose(spmd_result.noise_variance_, batch_result.noise_variance_, atol=tol)
+    assert_allclose(
+        spmd_result.explained_variance_ratio_,
+        batch_result.explained_variance_ratio_,
+        atol=tol,
+    )

sklearnex/spmd/ensemble/__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 .forest import RandomForestClassifier, RandomForestRegressor
+
+__all__ = ["RandomForestClassifier", "RandomForestRegressor"]

sklearnex/spmd/ensemble/forest.py

@@ -0,0 +1,71 @@
+# ==============================================================================
+# 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.ensemble import RandomForestClassifier as onedal_RandomForestClassifier
+from onedal.spmd.ensemble import RandomForestRegressor as onedal_RandomForestRegressor
+
+from ...ensemble import RandomForestClassifier as RandomForestClassifier_Batch
+from ...ensemble import RandomForestRegressor as RandomForestRegressor_Batch
+
+
+class RandomForestClassifier(RandomForestClassifier_Batch):
+    __doc__ = RandomForestClassifier_Batch.__doc__
+    _onedal_factory = onedal_RandomForestClassifier
+
+    def _onedal_cpu_supported(self, method_name, *data):
+        # TODO:
+        # check which methods supported SPMD interface on CPU.
+        ready = super()._onedal_cpu_supported(method_name, *data)
+        if not ready:
+            raise RuntimeError(
+                f"Method {method_name} in {self.__class__.__name__} "
+                "is not supported with given inputs."
+            )
+        return ready
+
+    def _onedal_gpu_supported(self, method_name, *data):
+        ready = super()._onedal_gpu_supported(method_name, *data)
+        if not ready:
+            raise RuntimeError(
+                f"Method {method_name} in {self.__class__.__name__} "
+                "is not supported with given inputs."
+            )
+        return ready
+
+
+class RandomForestRegressor(RandomForestRegressor_Batch):
+    __doc__ = RandomForestRegressor_Batch.__doc__
+    _onedal_factory = onedal_RandomForestRegressor
+
+    def _onedal_cpu_supported(self, method_name, *data):
+        # TODO:
+        # check which methods supported SPMD interface on CPU.
+        ready = super()._onedal_cpu_supported(method_name, *data)
+        if not ready:
+            raise RuntimeError(
+                f"Method {method_name} in {self.__class__.__name__} "
+                "is not supported with given inputs."
+            )
+        return ready
+
+    def _onedal_gpu_supported(self, method_name, *data):
+        ready = super()._onedal_gpu_supported(method_name, *data)
+        if not ready:
+            raise RuntimeError(
+                f"Method {method_name} in {self.__class__.__name__} "
+                "is not supported with given inputs."
+            )
+        return ready

sklearnex/spmd/ensemble/tests/test_forest_spmd.py

@@ -0,0 +1,265 @@
+# ==============================================================================
+# 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 sklearn.datasets import make_regression
+
+from onedal.tests.utils._dataframes_support import (
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex.tests.utils.spmd import (
+    _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_rfcls_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.ensemble import RandomForestClassifier as RandomForestClassifier_Batch
+    from sklearnex.spmd.ensemble import (
+        RandomForestClassifier as RandomForestClassifier_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, 2, 1, 2, 1, 0, 1, 2, 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 predictions of batch algo match spmd
+    spmd_model = RandomForestClassifier_SPMD(n_estimators=3, random_state=0).fit(
+        local_dpt_X_train, local_dpt_y_train
+    )
+    batch_model = RandomForestClassifier_Batch(n_estimators=3, random_state=0).fit(
+        X_train, y_train
+    )
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    pytest.skip("SPMD and batch random forest results not aligned")
+    _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, 1000])
+@pytest.mark.parametrize("n_features_and_classes", [(5, 2), (25, 2), (25, 10)])
+@pytest.mark.parametrize("n_estimators", [10, 100])
+@pytest.mark.parametrize("max_depth", [3, None])
+@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_rfcls_spmd_synthetic(
+    n_samples, n_features_and_classes, n_estimators, max_depth, dataframe, queue, dtype
+):
+    n_features, n_classes = n_features_and_classes
+    # Import spmd and batch algo
+    from sklearnex.ensemble import RandomForestClassifier as RandomForestClassifier_Batch
+    from sklearnex.spmd.ensemble import (
+        RandomForestClassifier as RandomForestClassifier_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 = RandomForestClassifier_SPMD(
+        n_estimators=n_estimators, max_depth=max_depth, random_state=0
+    ).fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = RandomForestClassifier_Batch(
+        n_estimators=n_estimators, max_depth=max_depth, random_state=0
+    ).fit(X_train, y_train)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    pytest.skip("SPMD and batch random forest results not aligned")
+    _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_rfreg_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.ensemble import RandomForestRegressor as RandomForestRegressor_Batch
+    from sklearnex.spmd.ensemble import (
+        RandomForestRegressor as RandomForestRegressor_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 predictions of batch algo match spmd
+    spmd_model = RandomForestRegressor_SPMD(n_estimators=3, random_state=0).fit(
+        local_dpt_X_train, local_dpt_y_train
+    )
+    batch_model = RandomForestRegressor_Batch(n_estimators=3, random_state=0).fit(
+        X_train, y_train
+    )
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    pytest.skip("SPMD and batch random forest results not aligned")
+    _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, 1000])
+@pytest.mark.parametrize("n_features", [5, 25])
+@pytest.mark.parametrize("n_estimators", [10, 100])
+@pytest.mark.parametrize("max_depth", [3, None])
+@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_rfreg_spmd_synthetic(
+    n_samples, n_features, n_estimators, max_depth, dataframe, queue, dtype
+):
+    # Import spmd and batch algo
+    from sklearnex.ensemble import RandomForestRegressor as RandomForestRegressor_Batch
+    from sklearnex.spmd.ensemble import (
+        RandomForestRegressor as RandomForestRegressor_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 = RandomForestRegressor_Batch(
+        n_estimators=n_estimators, max_depth=max_depth, random_state=0
+    ).fit(local_dpt_X_train, local_dpt_y_train)
+    batch_model = RandomForestRegressor_Batch(
+        n_estimators=n_estimators, max_depth=max_depth, random_state=0
+    ).fit(X_train, y_train)
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    # TODO: remove skips when SPMD and batch are aligned
+    pytest.skip("SPMD and batch random forest results not aligned")
+    _spmd_assert_allclose(spmd_result, batch_result)

sklearnex/spmd/linear_model/__init__.py

@@ -0,0 +1,21 @@
+# ==============================================================================
+# 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 .incremental_linear_model import IncrementalLinearRegression
+from .linear_model import LinearRegression
+from .logistic_regression import LogisticRegression
+
+__all__ = ["IncrementalLinearRegression", "LinearRegression", "LogisticRegression"]

sklearnex/spmd/linear_model/incremental_linear_model.py

@@ -0,0 +1,35 @@
+# ==============================================================================
+# 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 onedal.spmd.linear_model import (
+    IncrementalLinearRegression as onedalSPMD_IncrementalLinearRegression,
+)
+
+from ...linear_model import (
+    IncrementalLinearRegression as base_IncrementalLinearRegression,
+)
+
+
+class IncrementalLinearRegression(base_IncrementalLinearRegression):
+    """
+    Distributed incremental estimator for linear regression.
+    Allows for distributed training of linear regression if data is split into batches.
+
+    API is the same as for `sklearnex.linear_model.IncrementalLinearRegression`.
+    """
+
+    _onedal_incremental_linear = staticmethod(onedalSPMD_IncrementalLinearRegression)

sklearnex/spmd/linear_model/linear_model.py

@@ -0,0 +1,21 @@
+# ==============================================================================
+# 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.linear_model import LinearRegression
+
+# TODO:
+# Currently it uses `onedal` module interface.
+# Add sklearnex dispatching.

sklearnex/spmd/linear_model/logistic_regression.py

@@ -0,0 +1,21 @@
+# ==============================================================================
+# 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 onedal.spmd.linear_model import LogisticRegression
+
+# TODO:
+# Currently it uses `onedal` module interface.
+# Add sklearnex dispatching.