scikit-learn-intelex 2025.10.0__py313-none-win_amd64.whl

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/basic_statistics/tests/test_incremental_basic_statistics_spmd.py

@@ -0,0 +1,314 @@
+# ==============================================================================
+# 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.basic_statistics.tests.utils import options_and_tests
+from onedal.tests.utils._dataframes_support import (
+    _as_numpy,
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex import config_context
+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.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_incremental_basic_statistics_fit_spmd_gold(dataframe, queue, weighted, dtype):
+    # Import spmd and batch algo
+    from sklearnex.basic_statistics import IncrementalBasicStatistics
+    from sklearnex.spmd.basic_statistics import (
+        IncrementalBasicStatistics as IncrementalBasicStatistics_SPMD,
+    )
+
+    # Create gold data and process into dpt
+    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],
+        ],
+        dtype=dtype,
+    )
+    dpt_data = _convert_to_dataframe(data, sycl_queue=queue, target_df=dataframe)
+
+    local_dpt_data = _convert_to_dataframe(
+        _get_local_tensor(data), sycl_queue=queue, target_df=dataframe
+    )
+
+    if weighted:
+        # Create weights array containing the weight for each sample in the data
+        weights = np.array([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0], dtype=dtype)
+        dpt_weights = _convert_to_dataframe(
+            weights, sycl_queue=queue, target_df=dataframe
+        )
+        local_dpt_weights = _convert_to_dataframe(
+            _get_local_tensor(weights), sycl_queue=queue, target_df=dataframe
+        )
+
+    # ensure results of batch algo match spmd
+
+    incbs_spmd = IncrementalBasicStatistics_SPMD().fit(
+        local_dpt_data, sample_weight=local_dpt_weights if weighted else None
+    )
+    incbs = IncrementalBasicStatistics().fit(
+        dpt_data, sample_weight=dpt_weights if weighted else None
+    )
+
+    for option in options_and_tests:
+        attr = option + "_"
+        assert_allclose(
+            getattr(incbs_spmd, attr),
+            getattr(incbs, attr),
+            err_msg=f"Result for {option} is incorrect",
+        )
+
+
+@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.parametrize("num_blocks", [1, 2])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_incremental_basic_statistics_partial_fit_spmd_gold(
+    dataframe, queue, num_blocks, weighted, dtype
+):
+    # Import spmd and batch algo
+    from sklearnex.basic_statistics import IncrementalBasicStatistics
+    from sklearnex.spmd.basic_statistics import (
+        IncrementalBasicStatistics as IncrementalBasicStatistics_SPMD,
+    )
+
+    # Create gold data and process into dpt
+    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],
+        ],
+        dtype=dtype,
+    )
+    dpt_data = _convert_to_dataframe(data, sycl_queue=queue, target_df=dataframe)
+    local_data = _get_local_tensor(data)
+    split_local_data = np.array_split(local_data, num_blocks)
+
+    if weighted:
+        # Create weights array containing the weight for each sample in the data
+        weights = np.array([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0], dtype=dtype)
+        dpt_weights = _convert_to_dataframe(
+            weights, sycl_queue=queue, target_df=dataframe
+        )
+        local_weights = _get_local_tensor(weights)
+        split_local_weights = np.array_split(local_weights, num_blocks)
+
+    incbs_spmd = IncrementalBasicStatistics_SPMD()
+    incbs = IncrementalBasicStatistics()
+
+    for i in range(num_blocks):
+        local_dpt_data = _convert_to_dataframe(
+            split_local_data[i], sycl_queue=queue, target_df=dataframe
+        )
+        if weighted:
+            local_dpt_weights = _convert_to_dataframe(
+                split_local_weights[i], sycl_queue=queue, target_df=dataframe
+            )
+        incbs_spmd.partial_fit(
+            local_dpt_data, sample_weight=local_dpt_weights if weighted else None
+        )
+
+    incbs.fit(dpt_data, sample_weight=dpt_weights if weighted else None)
+
+    for option in options_and_tests:
+        attr = option + "_"
+        assert_allclose(
+            getattr(incbs_spmd, attr),
+            getattr(incbs, attr),
+            err_msg=f"Result for {option} is incorrect",
+        )
+
+
+@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.parametrize("num_blocks", [1, 2])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("option", options_and_tests.keys())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.mpi
+def test_incremental_basic_statistics_single_option_partial_fit_spmd_gold(
+    dataframe, queue, num_blocks, weighted, option, dtype
+):
+    # Import spmd and batch algo
+    from sklearnex.basic_statistics import IncrementalBasicStatistics
+    from sklearnex.spmd.basic_statistics import (
+        IncrementalBasicStatistics as IncrementalBasicStatistics_SPMD,
+    )
+
+    # Create gold data and process into dpt
+    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],
+        ],
+        dtype=dtype,
+    )
+    dpt_data = _convert_to_dataframe(data, sycl_queue=queue, target_df=dataframe)
+    local_data = _get_local_tensor(data)
+    split_local_data = np.array_split(local_data, num_blocks)
+
+    if weighted:
+        # Create weights array containing the weight for each sample in the data
+        weights = np.array([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0], dtype=dtype)
+        dpt_weights = _convert_to_dataframe(
+            weights, sycl_queue=queue, target_df=dataframe
+        )
+        local_weights = _get_local_tensor(weights)
+        split_local_weights = np.array_split(local_weights, num_blocks)
+
+    incbs_spmd = IncrementalBasicStatistics_SPMD(result_options=option)
+    incbs = IncrementalBasicStatistics(result_options=option)
+
+    for i in range(num_blocks):
+        local_dpt_data = _convert_to_dataframe(
+            split_local_data[i], sycl_queue=queue, target_df=dataframe
+        )
+        if weighted:
+            local_dpt_weights = _convert_to_dataframe(
+                split_local_weights[i], sycl_queue=queue, target_df=dataframe
+            )
+        incbs_spmd.partial_fit(
+            local_dpt_data, sample_weight=local_dpt_weights if weighted else None
+        )
+
+    incbs.fit(dpt_data, sample_weight=dpt_weights if weighted else None)
+    attr = option + "_"
+    assert_allclose(getattr(incbs_spmd, attr), getattr(incbs, attr))
+
+
+@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.parametrize("num_blocks", [1, 2])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("n_samples", [100, 10000])
+@pytest.mark.parametrize("n_features", [10, 100])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+@pytest.mark.parametrize("use_raw_input", [True, False])
+@pytest.mark.mpi
+def test_incremental_basic_statistics_partial_fit_spmd_synthetic(
+    dataframe, queue, num_blocks, weighted, n_samples, n_features, dtype, use_raw_input
+):
+    # Import spmd and batch algo
+    from sklearnex.basic_statistics import IncrementalBasicStatistics
+    from sklearnex.spmd.basic_statistics import (
+        IncrementalBasicStatistics as IncrementalBasicStatistics_SPMD,
+    )
+
+    tol = 2e-3 if dtype == np.float32 else 1e-7
+
+    # Create gold data and process into dpt
+    data = _generate_statistic_data(n_samples, n_features, dtype=dtype)
+    local_data = _get_local_tensor(data)
+    split_local_data = np.array_split(local_data, num_blocks)
+    split_data = np.array_split(data, num_blocks)
+
+    if weighted:
+        # Create weights array containing the weight for each sample in the data
+        weights = _generate_statistic_data(n_samples, dtype=dtype)
+        local_weights = _get_local_tensor(weights)
+        split_local_weights = np.array_split(local_weights, num_blocks)
+        split_weights = np.array_split(weights, num_blocks)
+
+    incbs_spmd = IncrementalBasicStatistics_SPMD()
+    incbs = IncrementalBasicStatistics()
+
+    for i in range(num_blocks):
+        local_dpt_data = _convert_to_dataframe(
+            split_local_data[i], sycl_queue=queue, target_df=dataframe
+        )
+        dpt_data = _convert_to_dataframe(
+            split_data[i], sycl_queue=queue, target_df=dataframe
+        )
+        if weighted:
+            local_dpt_weights = _convert_to_dataframe(
+                split_local_weights[i], sycl_queue=queue, target_df=dataframe
+            )
+            dpt_weights = _convert_to_dataframe(
+                split_weights[i], sycl_queue=queue, target_df=dataframe
+            )
+        # Configure raw input status for spmd estimator
+        with config_context(use_raw_input=use_raw_input):
+            incbs_spmd.partial_fit(
+                local_dpt_data, sample_weight=local_dpt_weights if weighted else None
+            )
+        incbs.partial_fit(dpt_data, sample_weight=dpt_weights if weighted else None)
+
+    for option in options_and_tests:
+        attr = option + "_"
+        assert_allclose(
+            _as_numpy(getattr(incbs_spmd, attr)),
+            _as_numpy(getattr(incbs, attr)),
+            atol=tol,
+            err_msg=f"Result for {option} is incorrect",
+        )

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/cluster/__init__.py

@@ -0,0 +1,30 @@
+# ==============================================================================
+# 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 daal4py.sklearn._utils import daal_check_version
+
+from .dbscan import DBSCAN
+
+if daal_check_version((2023, "P", 200)):
+    from .kmeans import KMeans
+
+    __all__ = ["DBSCAN", "KMeans"]
+else:
+    # TODO:
+    # update versioning for DBSCAN.
+    __all__ = [
+        "DBSCAN",
+    ]

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/cluster/dbscan.py

@@ -0,0 +1,26 @@
+# ==============================================================================
+# 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 abc import ABC
+
+from onedal.spmd.cluster import DBSCAN as onedal_DBSCAN
+
+from ...cluster import DBSCAN as DBSCAN_Batch
+
+
+class DBSCAN(DBSCAN_Batch):
+    __doc__ = DBSCAN_Batch.__doc__
+    _onedal_dbscan = staticmethod(onedal_DBSCAN)

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/cluster/kmeans.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.cluster import KMeans
+
+# TODO:
+# Currently it uses `onedal` module interface.
+# Add sklearnex dispatching.

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/cluster/tests/test_dbscan_spmd.py

@@ -0,0 +1,108 @@
+# ==============================================================================
+# 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 onedal.tests.utils._dataframes_support import (
+    _convert_to_dataframe,
+    get_dataframes_and_queues,
+)
+from sklearnex import config_context
+from sklearnex.tests.utils.spmd import (
+    _generate_clustering_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_dbscan_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.cluster import DBSCAN as DBSCAN_Batch
+    from sklearnex.spmd.cluster import DBSCAN as DBSCAN_SPMD
+
+    data = np.array([[1, 2], [2, 2], [2, 3], [8, 7], [8, 8], [25, 80]])
+
+    local_dpt_data = _convert_to_dataframe(
+        _get_local_tensor(data), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure labels from fit of batch algo matches spmd
+    spmd_model = DBSCAN_SPMD(eps=3, min_samples=2).fit(local_dpt_data)
+    batch_model = DBSCAN_Batch(eps=3, min_samples=2).fit(data)
+
+    _spmd_assert_allclose(spmd_model.labels_, batch_model.labels_)
+
+
+@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_eps", [(5, 3), (5, 10), (25, 10)])
+@pytest.mark.parametrize("centers", [10, None])
+@pytest.mark.parametrize("min_samples", [2, 5, 15])
+@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.parametrize("use_raw_input", [True, False])
+@pytest.mark.mpi
+def test_dbscan_spmd_synthetic(
+    n_samples,
+    n_features_and_eps,
+    centers,
+    min_samples,
+    dataframe,
+    queue,
+    dtype,
+    use_raw_input,
+):
+    n_features, eps = n_features_and_eps
+    # Import spmd and batch algo
+    from sklearnex.cluster import DBSCAN as DBSCAN_Batch
+    from sklearnex.spmd.cluster import DBSCAN as DBSCAN_SPMD
+
+    data, _ = _generate_clustering_data(
+        n_samples, n_features, centers=centers, dtype=dtype
+    )
+
+    local_dpt_data = _convert_to_dataframe(
+        _get_local_tensor(data), sycl_queue=queue, target_df=dataframe
+    )
+
+    # Ensure labels from fit of batch algo matches spmd
+    # Configure raw input status for spmd estimator
+    with config_context(use_raw_input=use_raw_input):
+        spmd_model = DBSCAN_SPMD(eps=eps, min_samples=min_samples).fit(local_dpt_data)
+    batch_model = DBSCAN_Batch(eps=eps, min_samples=min_samples).fit(data)
+
+    _spmd_assert_allclose(spmd_model.labels_, batch_model.labels_)
+
+    # Ensure meaningful test setup
+    if np.all(batch_model.labels_ == -1):
+        raise ValueError("No labels given - try raising epsilon")

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/sklearnex/spmd/cluster/tests/test_kmeans_spmd.py

@@ -0,0 +1,180 @@
+# ==============================================================================
+# 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 import config_context
+from sklearnex.tests.utils.spmd import (
+    _assert_kmeans_labels_allclose,
+    _assert_unordered_allclose,
+    _generate_clustering_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_kmeans_spmd_gold(dataframe, queue):
+    # Import spmd and batch algo
+    from sklearnex.cluster import KMeans as KMeans_Batch
+    from sklearnex.spmd.cluster import KMeans as KMeans_SPMD
+
+    X_train = np.array(
+        [
+            [1, 2],
+            [2, 2],
+            [2, 3],
+            [8, 7],
+            [8, 8],
+            [25, 80],
+            [5, 65],
+            [2, 8],
+            [1, 3],
+            [2, 2],
+            [1, 3],
+            [2, 2],
+        ],
+        dtype=np.float64,
+    )
+    X_test = np.array([[0, 0], [12, 3], [2, 2], [7, 8]], dtype=np.float64)
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_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 labels from fit of batch algo matches spmd
+    spmd_model = KMeans_SPMD(n_clusters=2, random_state=0).fit(local_dpt_X_train)
+    batch_model = KMeans_Batch(n_clusters=2, random_state=0).fit(X_train)
+
+    _assert_unordered_allclose(spmd_model.cluster_centers_, batch_model.cluster_centers_)
+    _assert_kmeans_labels_allclose(
+        spmd_model.labels_,
+        batch_model.labels_,
+        spmd_model.cluster_centers_,
+        batch_model.cluster_centers_,
+    )
+    assert_allclose(spmd_model.n_iter_, batch_model.n_iter_, atol=1)
+
+    # Ensure predictions of batch algo match spmd
+    spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _assert_kmeans_labels_allclose(
+        spmd_result,
+        batch_result,
+        spmd_model.cluster_centers_,
+        batch_model.cluster_centers_,
+    )
+
+
+@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_clusters", [2, 5, 15])
+@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.parametrize("use_raw_input", [True, False])
+@pytest.mark.mpi
+def test_kmeans_spmd_synthetic(
+    n_samples, n_features, n_clusters, dataframe, queue, dtype, use_raw_input
+):
+    # Import spmd and batch algo
+    from sklearnex.cluster import KMeans as KMeans_Batch
+    from sklearnex.spmd.cluster import KMeans as KMeans_SPMD
+
+    # TODO: investigate issues when centers != n_clusters (spmd and batch results don't match for all values of K)
+    X_train, X_test = _generate_clustering_data(
+        n_samples, n_features, centers=n_clusters, dtype=dtype
+    )
+
+    local_dpt_X_train = _convert_to_dataframe(
+        _get_local_tensor(X_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
+    )
+
+    # Validate KMeans init
+    spmd_model_init = KMeans_SPMD(n_clusters=n_clusters, max_iter=1, random_state=0).fit(
+        local_dpt_X_train
+    )
+    batch_model_init = KMeans_Batch(
+        n_clusters=n_clusters, max_iter=1, random_state=0
+    ).fit(X_train)
+    # TODO: centers do not match up after init
+    # _assert_unordered_allclose(spmd_model_init.cluster_centers_, batch_model_init.cluster_centers_)
+
+    # Ensure labels from fit of batch algo matches spmd, using same init
+    spmd_model = KMeans_SPMD(
+        n_clusters=n_clusters, init=spmd_model_init.cluster_centers_, random_state=0
+    )
+    # Configure raw input status for spmd estimator
+    with config_context(use_raw_input=use_raw_input):
+        spmd_model.fit(local_dpt_X_train)
+    batch_model = KMeans_Batch(
+        n_clusters=n_clusters, init=spmd_model_init.cluster_centers_, random_state=0
+    ).fit(X_train)
+
+    atol = 1e-5 if dtype == np.float32 else 1e-7
+    _assert_unordered_allclose(
+        spmd_model.cluster_centers_, batch_model.cluster_centers_, atol=atol
+    )
+    _assert_kmeans_labels_allclose(
+        spmd_model.labels_,
+        batch_model.labels_,
+        spmd_model.cluster_centers_,
+        batch_model.cluster_centers_,
+        atol=atol,
+    )
+    # TODO: KMeans iterations are not aligned
+    # assert_allclose(spmd_model.n_iter_, batch_model.n_iter_, atol=1)
+
+    # Ensure predictions of batch algo match spmd
+    # Configure raw input status for spmd estimator
+    with config_context(use_raw_input=use_raw_input):
+        spmd_result = spmd_model.predict(local_dpt_X_test)
+    batch_result = batch_model.predict(X_test)
+
+    _assert_kmeans_labels_allclose(
+        spmd_result,
+        batch_result,
+        spmd_model.cluster_centers_,
+        batch_model.cluster_centers_,
+        atol=atol,
+    )