scikit-learn-intelex 2025.6.1__py311-none-win_amd64.whl → 2025.8.0__py311-none-win_amd64.whl

scikit_learn_intelex-2025.8.0.data/data/Lib/site-packages/onedal/datatypes/_data_conversion.py

@@ -0,0 +1,156 @@
+# ==============================================================================
+# 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 scipy.sparse as sp
+
+from onedal import _default_backend as backend
+
+from ..utils._third_party import is_dpctl_tensor, is_dpnp_ndarray, lazy_import
+
+
+def _apply_and_pass(func, *args, **kwargs):
+    if len(args) == 1:
+        return func(args[0], **kwargs)
+    return tuple(map(lambda arg: func(arg, **kwargs), args))
+
+
+def _convert_one_to_table(arg, queue=None):
+    # All inputs for table conversion must be array-like or sparse, not scalars
+    return backend.to_table(np.atleast_2d(arg) if np.isscalar(arg) else arg, queue)
+
+
+def to_table(*args, queue=None):
+    """Create oneDAL tables from scalars and/or arrays.
+
+    Parameters
+    ----------
+    *args : scalar, numpy array, sycl_usm_ndarray, csr_matrix, or csr_array
+        Arguments to be individually converted to oneDAL tables.
+
+    queue : SyclQueue or None, default=None
+        SYCL Queue object to be associated with the oneDAL tables. Default
+        value None causes no change in data location or queue.
+
+    Returns
+    -------
+    tables : oneDAL homogeneous or csr tables
+        Converted oneDAL tables for scalar, numpy array, sycl_usm_ndarray
+        and dlpack inputs; oneDAL csr_table for csr_matrix, csr_array.
+
+    Notes
+    -----
+        Tables will use pointers to the original array data. Scalars
+        and non-contiguous arrays will be copies. Arrays may be
+        modified in-place by oneDAL during computation. Transformation
+        is possible only for data located on CPU and SYCL-enabled Intel
+        GPUs. Each array may only be of a single data type (i.e. each
+        must be homogeneous).
+    """
+    return _apply_and_pass(_convert_one_to_table, *args, queue=queue)
+
+
+@lazy_import("array_api_compat")
+def _compat_convert(array_api_compat, array):
+    return array_api_compat.get_namespace(array).from_dlpack
+
+
+def return_type_constructor(array):
+    """Generate a function for converting oneDAL tables to arrays.
+
+    Parameters
+    ----------
+    array : array-like or None
+        Python object representing an array instance of the return type
+        for converting oneDAL tables. Arrays are queried for conversion
+        namespace when of sycl_usm_array type or array API standard type.
+        When set to None, will return numpy arrays or scipy csr arrays.
+
+    Returns
+    -------
+    func : callable
+        A function which takes in a single table input and returns an array.
+    """
+    if isinstance(array, np.ndarray) or array is None or sp.issparse(array):
+        func = backend.from_table
+    elif hasattr(array, "__sycl_usm_array_interface__"):
+        # oneDAL returns tables without sycl queues for CPU sycl queue inputs.
+        # This workaround is necessary for the functional preservation
+        # of the compute-follows-data execution.
+        device = array.sycl_queue
+        # Its important to note why the __sycl_usm_array_interface__ is
+        # prioritized: it provides finer-grained control of SYCL queues and the
+        # related SYCL devices which are generally unavailable via DLPack
+        # representations (such as SYCL contexts, SYCL sub-devices, etc.).
+        xp = array.__array_namespace__()
+        # array api support added in dpnp starting in 0.19, will fail for
+        # older versions
+        if is_dpctl_tensor(array):
+            func = lambda x: (
+                xp.asarray(x)
+                if hasattr(x, "__sycl_usm_array_interface__")
+                else xp.asarray(backend.from_table(x), device=device)
+            )
+        elif is_dpnp_ndarray(array):
+            func = lambda x: (
+                xp.asarray(xp.as_usm_ndarray(x))
+                if hasattr(x, "__sycl_usm_array_interface__")
+                else xp.asarray(backend.from_table(x), device=device)
+            )
+
+    elif hasattr(array, "__array_namespace__"):
+        func = array.__array_namespace__().from_dlpack
+    else:
+        try:
+            func = _compat_convert(array)
+        except ImportError:
+            raise TypeError(
+                "array type is unsupported, but may be made compatible by installing `array_api_compat`"
+            ) from None
+    return func
+
+
+def from_table(*args, like=None):
+    """Create 2 dimensional arrays from oneDAL tables.
+
+    oneDAL tables are converted to numpy ndarrays, dpctl tensors, dpnp
+    ndarrays, or array API standard arrays of designated type.
+
+    Parameters
+    ----------
+    *args : single or multiple python oneDAL tables
+        The tables should given as individual arguments.
+
+    like : callable, array-like or None, default=None
+        Python object representing an array instance of the return type
+        or function capable of converting oneDAL tables into arrays of
+        desired type. Arrays are queried for conversion namespace when
+        of sycl_usm_array type or array API standard type. When set to
+        None, will return numpy arrays or scipy csr arrays.
+
+    Returns
+    -------
+    arrays : numpy arrays, sycl_usm_ndarrays, or array API standard arrays
+        Array or tuple of arrays generated from the input oneDAL tables.
+
+    Notes
+    -----
+    Support for other array types via array_api_compat is possible (e.g.
+    PyTorch), but requires its installation specifically, as it is imported
+    only when necessary.
+    """
+    func = like if callable(like) else return_type_constructor(like)
+    return _apply_and_pass(func, *args)

scikit_learn_intelex-2025.8.0.data/data/Lib/site-packages/onedal/datatypes/_dlpack.py

@@ -0,0 +1,61 @@
+# ==============================================================================
+# Copyright Contributors to the oneDAL Project
+#
+# 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 collections.abc import Iterable
+
+import numpy as np
+
+from onedal import _default_backend as backend
+
+from ..utils._third_party import convert_sklearnex_queue, lazy_import
+
+cpu_dlpack_device = (backend.kDLCPU, 0)
+
+
+@lazy_import("torch.xpu")
+@convert_sklearnex_queue
+def get_torch_queue(torchxpu, array):
+    return backend.SyclQueue(torchxpu.current_stream(array.get_device()).sycl_queue)
+
+
+def dlpack_to_numpy(obj):
+    # check dlpack data location.
+    if obj.__dlpack_device__() != cpu_dlpack_device:
+        if hasattr(obj, "to_device"):
+            # use of the "cpu" string as device not officially part of
+            # the array api standard but widely supported
+            obj = obj.to_device("cpu")
+        elif hasattr(obj, "to"):
+            # pytorch-specific fix as it is not array api compliant
+            obj = obj.to("cpu")
+        else:
+            raise TypeError(f"cannot move {type(obj)} to cpu")
+
+    # convert to numpy
+    if hasattr(obj, "__array__"):
+        # ``copy`` param for the ``asarray`` is not set.
+        # The object is copied only if needed
+        obj = np.asarray(obj)
+    else:
+        # requires numpy 1.23
+        try:
+            obj = np.from_dlpack(obj)
+        except AttributeError:
+            raise NotImplementedError(
+                "Upgrade NumPy >= 1.23 for dlpack support"
+            ) from None
+
+    return obj

scikit_learn_intelex-2025.8.0.data/data/Lib/site-packages/onedal/datatypes/_sycl_usm.py

@@ -0,0 +1,63 @@
+# ==============================================================================
+# Copyright Contributors to the oneDAL Project
+#
+# 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 collections.abc import Iterable
+
+import numpy as np
+import scipy.sparse as sp
+
+from ..utils._third_party import lazy_import
+
+
+@lazy_import("dpctl.memory", "dpctl.tensor")
+def _array_to_usm(memory, tensor, queue, array):
+    try:
+        mem = memory.MemoryUSMDevice(array.nbytes, queue=queue)
+        mem.copy_from_host(array.tobytes())
+        return tensor.usm_ndarray(array.shape, array.dtype, buffer=mem)
+    except ValueError as e:
+        # ValueError will raise if device does not support the dtype
+        # retry with float32 (needed for fp16 and fp64 support issues)
+        # try again as float32, if it is a float32 just raise the error.
+        if array.dtype == np.float32:
+            raise e
+        return _array_to_usm(queue, array.astype(np.float32))
+
+
+@lazy_import("dpnp", "dpctl.tensor")
+def _to_dpnp(dpnp, tensor, array):
+    if isinstance(array, tensor.usm_ndarray):
+        return dpnp.array(array, copy=False)
+    else:
+        return array
+
+
+def copy_to_usm(queue, array):
+    if hasattr(array, "tobytes"):
+        return _array_to_usm(queue, array)
+    else:
+        if isinstance(array, Iterable) and not sp.issparse(array):
+            array = [copy_to_usm(queue, i) for i in array]
+        return array
+
+
+def copy_to_dpnp(queue, array):
+    if hasattr(array, "tobytes"):
+        return _to_dpnp(_array_to_usm(queue, array))
+    else:
+        if isinstance(array, Iterable) and not sp.issparse(array):
+            array = [copy_to_dpnp(queue, i) for i in array]
+        return array

{scikit_learn_intelex-2025.6.1.data → scikit_learn_intelex-2025.8.0.data}/data/Lib/site-packages/onedal/datatypes/tests/common.py

@@ -14,11 +14,16 @@
 # limitations under the License.
 # ===============================================================================
 
-from onedal.utils._dpep_helpers import dpctl_available, dpnp_available
+from onedal.utils._third_party import dpctl_available
 
-if dpnp_available:
+try:
     import dpnp
 
+    dpnp_available = True
+except ImportError:
+    dpnp_available = False
+
+
 if dpctl_available:
     import dpctl
     from dpctl.tensor import usm_ndarray
@@ -99,7 +104,7 @@ if dpctl_available:
         if not skip_data_1:
             assert actual_sua_iface["data"][1] == desired_sua_iface["data"][1]
         # shape: a tuple of integers describing dimensions of an N-dimensional array.
-        # Reformating shapes for check cases (r,) vs (r,1). Contiguous flattened array
+        # Reformatting shapes for check cases (r,) vs (r,1). Contiguous flattened array
         # shape (r,) becoming (r,1) just for the check, since oneDAL supports only (r,1)
         # for 1-D arrays. In code after from_table conversion for 1-D expected outputs
         # xp.ravel or reshape(-1) is used.

{scikit_learn_intelex-2025.6.1.data → scikit_learn_intelex-2025.8.0.data}/data/Lib/site-packages/onedal/datatypes/tests/test_data.py

@@ -14,15 +14,17 @@
 # limitations under the License.
 # ==============================================================================
 
+import gc
+
 import numpy as np
 import pytest
 import scipy.sparse as sp
-from numpy.testing import assert_allclose
+from numpy.testing import assert_allclose, assert_array_equal
 
 from onedal import _default_backend, _dpc_backend
 from onedal._device_offload import supports_queue
 from onedal.datatypes import from_table, to_table
-from onedal.utils._dpep_helpers import dpctl_available
+from onedal.utils._third_party import dpctl_available
 
 backend = _dpc_backend or _default_backend
 
@@ -79,12 +81,7 @@ class DummyEstimatorWithTableConversions:
         # oneDAL backend func is needed to check result table checks.
         result = dbscan.compute(params, X_table, to_table(None))
         result_responses_table = result.responses
-        result_responses_df = from_table(
-            result_responses_table,
-            sua_iface=sua_iface,
-            sycl_queue=X.sycl_queue,
-            xp=xp,
-        )
+        result_responses_df = from_table(result_responses_table, like=X)
         return X_table, result_responses_table, result_responses_df
 
 
@@ -270,9 +267,7 @@ def test_input_zero_copy_sycl_usm(dataframe, queue, order, dtype):
     X_table = to_table(X_dp)
     _assert_sua_iface_fields(X_dp, X_table)
 
-    X_dp_from_table = from_table(
-        X_table, sycl_queue=queue, sua_iface=sua_iface, xp=X_dp_namespace
-    )
+    X_dp_from_table = from_table(X_table, like=X_dp)
     _assert_sua_iface_fields(X_table, X_dp_from_table)
     _assert_tensor_attr(X_dp, X_dp_from_table, order)
 
@@ -305,8 +300,13 @@ def test_table_conversions_sycl_usm(dataframe, queue, order, data_shape, dtype):
     alg = DummyEstimatorWithTableConversions()
     X_table, result_responses_table, result_responses_df = alg.fit(X)
 
-    for obj in [X_table, result_responses_table, result_responses_df, X]:
+    for obj in [X_table, result_responses_df, X]:
         assert hasattr(obj, "__sycl_usm_array_interface__"), f"{obj} has no SUA interface"
+
+    assert (
+        hasattr(result_responses_table, "__sycl_usm_array_interface__")
+        != queue.sycl_device.is_cpu
+    )
     _assert_sua_iface_fields(X, X_table)
 
     # Work around for saving compute-follows-data execution
@@ -320,13 +320,14 @@ def test_table_conversions_sycl_usm(dataframe, queue, order, data_shape, dtype):
     # after conversion from onedal table to sua array.
     # Test is not turned off because of this. Only check is skipped.
     skip_data_1 = True
-    _assert_sua_iface_fields(
-        result_responses_df,
-        result_responses_table,
-        skip_data_0=skip_data_0,
-        skip_data_1=skip_data_1,
-        skip_syclobj=skip_syclobj,
-    )
+    if not queue.sycl_device.is_cpu:
+        _assert_sua_iface_fields(
+            result_responses_df,
+            result_responses_table,
+            skip_data_0=skip_data_0,
+            skip_data_1=skip_data_1,
+            skip_syclobj=skip_syclobj,
+        )
     assert X.sycl_queue == result_responses_df.sycl_queue
     if order == "F":
         assert X.flags.f_contiguous == result_responses_df.flags.f_contiguous
@@ -574,3 +575,44 @@ def test_table_conversions_dlpack(dataframe, queue, order, data_shape, dtype):
     # oneDAL table construction sets 1d arrays to 2d arrays with 1 col
     # this is counter the numpy strategy, and requires numpy's squeeze
     assert_allclose(np.squeeze(X), np.squeeze(X_out))
+
+
+@pytest.mark.parametrize(
+    "dataframe,queue", get_dataframes_and_queues("dpctl,numpy,array_api", "cpu,gpu")
+)
+@pytest.mark.parametrize("order", ["F", "C"])
+@pytest.mark.parametrize("data_shape", data_shapes)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
+def test_table___dlpack__(dataframe, queue, order, data_shape, dtype):
+    """Test if __dlpack__ attribute can be properly consumed by other frameworks
+    This tests kDLOneAPI devices as well as kDLCPU devices.
+    """
+    rng = np.random.RandomState(0)
+    X = np.array(5 * rng.random_sample(data_shape), dtype=dtype)
+
+    X = ORDER_DICT[order](X)
+
+    X_df = _convert_to_dataframe(X, sycl_queue=queue, target_df=dataframe)
+
+    X_table = to_table(X_df)
+    if hasattr(X_df, "__dlpack_device__"):
+        assert X_df.__dlpack_device__() == X_table.__dlpack_device__()
+
+    if xp := getattr(X_df, "__array_namespace__", lambda: None)():
+        X_out = xp.from_dlpack(X_table)
+        X_temp = xp.asnumpy(X_out) if hasattr(xp, "asnumpy") else np.asarray(X)
+        if X_temp.dtype == X.dtype:
+            assert_array_equal(np.squeeze(X_temp), np.squeeze(X))
+        else:
+            assert_allclose(np.squeeze(X_temp), np.squeeze(X))
+    else:
+        # only some numpy versions support array_api and from_dlpack
+        pytest.skip(f"{dataframe} does not have an __array_namespace__ attribute")
+
+    # test capsule deletion, should have no impact on underlying memory
+    # important for testing ``dlpack::free_capsule``
+    capsule = X_table.__dlpack__()
+    assert_allclose(np.squeeze(from_table(X_table)), np.squeeze(X))
+    del capsule
+    gc.collect()
+    assert_allclose(np.squeeze(from_table(X_table)), np.squeeze(X))

{scikit_learn_intelex-2025.6.1.data → scikit_learn_intelex-2025.8.0.data}/data/Lib/site-packages/onedal/decomposition/incremental_pca.py

@@ -28,9 +28,7 @@ from .pca import BasePCA
 
 
 class IncrementalPCA(BasePCA):
-    """
-    Incremental estimator for PCA based on oneDAL implementation.
-    Allows to compute PCA if data are splitted into batches.
+    """Incremental oneDAL PCA estimator.
 
     Parameters
     ----------
@@ -42,8 +40,8 @@ class IncrementalPCA(BasePCA):
         When True the ``components_`` vectors are chosen in deterministic
         way, otherwise some of them can be oppositely directed.
 
-    method : string, default='cov'
-        Method used on oneDAL side to compute result.
+    method : str, default='cov'
+        Method used in oneDAL computation.
 
     whiten : bool, default=False
         When True the ``components_`` vectors are divided
@@ -86,7 +84,6 @@ class IncrementalPCA(BasePCA):
         noise_variance_ : float
             Equal to the average of (min(n_features, n_samples) - n_components)
             smallest eigenvalues of the covariance matrix of X.
-
     """
 
     def __init__(
@@ -131,7 +128,7 @@ class IncrementalPCA(BasePCA):
 
     @supports_queue
     def partial_fit(self, X, queue=None):
-        """Incremental fit with X. All of X is processed as a single batch.
+        """Generate partial PCA from batch data in `_partial_result`.
 
         Parameters
         ----------
@@ -139,8 +136,9 @@ class IncrementalPCA(BasePCA):
             Training data, where `n_samples` is the number of samples and
             `n_features` is the number of features.
 
-        y : Ignored
-            Not used, present for API consistency by convention.
+        queue : SyclQueue or None, default=None
+            SYCL Queue object for device code execution. Default
+            value None causes computation on host.
 
         Returns
         -------
@@ -190,14 +188,7 @@ class IncrementalPCA(BasePCA):
         return self
 
     def finalize_fit(self):
-        """
-        Finalizes principal components computation and obtains resulting
-        attributes from the current `_partial_result`.
-
-        Parameters
-        ----------
-        queue : dpctl.SyclQueue
-            Not used here, added for API conformance
+        """Finalize statistics from the current `_partial_result`.
 
         Returns
         -------

{scikit_learn_intelex-2025.6.1.data → scikit_learn_intelex-2025.8.0.data}/data/Lib/site-packages/onedal/decomposition/pca.py

@@ -25,14 +25,12 @@ from onedal._device_offload import supports_queue
 from onedal.common._backend import bind_default_backend
 
 from .._config import _get_config
+from ..common.hyperparameters import get_hyperparameters
 from ..datatypes import from_table, to_table
 from ..utils._array_api import _get_sycl_namespace
 
 
 class BasePCA(metaclass=ABCMeta):
-    """
-    Base class for PCA oneDAL implementation.
-    """
 
     def __init__(
         self,
@@ -172,7 +170,11 @@ class PCA(BasePCA):
 
         X = to_table(X, queue=queue)
         params = self._get_onedal_params(X)
-        result = self.train(params, X)
+        hparams = get_hyperparameters("pca", "train")
+        if hparams is not None and not hparams.is_default:
+            result = self.train(params, hparams.backend, X)
+        else:
+            result = self.train(params, X)
 
         self.mean_ = from_table(result.means).ravel()
         self.variances_ = from_table(result.variances)

{scikit_learn_intelex-2025.6.1.data → scikit_learn_intelex-2025.8.0.data}/data/Lib/site-packages/onedal/ensemble/forest.py

@@ -34,7 +34,6 @@ from ..common._estimator_checks import _check_is_fitted
 from ..common._mixin import ClassifierMixin, RegressorMixin
 from ..datatypes import from_table, to_table
 from ..utils._array_api import _get_sycl_namespace
-from ..utils._dpep_helpers import get_unique_values_with_dpep
 from ..utils.validation import (
     _check_array,
     _check_n_features,
@@ -315,7 +314,13 @@ class BaseForest(BaseEnsemble, metaclass=ABCMeta):
         else:
             if sua_iface is not None:
                 queue = X.sycl_queue
-            self.classes_ = get_unique_values_with_dpep(y)
+            # try catch needed for raw_inputs + array_api data where unlike
+            # numpy the way to yield unique values is via `unique_values`
+            # This should be removed when refactored for gpu zero-copy
+            try:
+                self.classes_ = xp.unique(y)
+            except AttributeError:
+                self.classes_ = xp.unique_values(y)
 
         self.n_features_in_ = X.shape[1]
 
@@ -361,7 +366,7 @@ class BaseForest(BaseEnsemble, metaclass=ABCMeta):
 
     def _create_model(self, module):
         # TODO:
-        # upate error msg.
+        # update error msg.
         raise NotImplementedError("Creating model is not supported.")
 
     def _predict(self, X, hparams=None):
@@ -385,14 +390,14 @@ class BaseForest(BaseEnsemble, metaclass=ABCMeta):
 
         model = self._onedal_model
         queue = QM.get_global_queue()
-        X = to_table(X, queue=queue)
-        params = self._get_onedal_params(X)
+        X_table = to_table(X, queue=queue)
+        params = self._get_onedal_params(X_table)
         if hparams is not None and not hparams.is_default:
-            result = self.infer(params, hparams.backend, model, X)
+            result = self.infer(params, hparams.backend, model, X_table)
         else:
-            result = self.infer(params, model, X)
+            result = self.infer(params, model, X_table)
 
-        y = from_table(result.responses, sua_iface=sua_iface, sycl_queue=queue, xp=xp)
+        y = from_table(result.responses, like=X)
         return y
 
     def _predict_proba(self, X, hparams=None):
@@ -517,7 +522,8 @@ class RandomForestClassifier(ClassifierMixin, BaseForest, metaclass=ABCMeta):
 
         try:
             return xp.take(
-                xp.asarray(self.classes_), xp.astype(xp.reshape(pred, (-1,)), xp.int64)
+                xp.asarray(self.classes_, device=pred.sycl_queue),
+                xp.astype(xp.reshape(pred, (-1,)), xp.int64),
             )
         except AttributeError:
             return np.take(self.classes_, pred.ravel().astype(np.int64, casting="unsafe"))