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

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/onedal/basic_statistics/incremental_basic_statistics.py

@@ -0,0 +1,165 @@
+# 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 .._device_offload import supports_queue
+from ..common._backend import bind_default_backend
+from ..datatypes import from_table, return_type_constructor, to_table
+from ..utils import _sycl_queue_manager as QM
+from .basic_statistics import BasicStatistics
+
+
+class IncrementalBasicStatistics(BasicStatistics):
+    """Incremental oneDAL low order moments estimator.
+
+    Calculate basic statistics for data split into batches.
+
+    Parameters
+    ----------
+    result_options : str or list, default=str('all')
+        List of statistics to compute.
+
+    algorithm : str, default=str('by_default')
+        Method for statistics computation.
+
+    Attributes
+    ----------
+        min : ndarray of shape (n_features,)
+            Minimum of each feature over all samples.
+
+        max : ndarray of shape (n_features,)
+            Maximum of each feature over all samples.
+
+        sum : ndarray of shape (n_features,)
+            Sum of each feature over all samples.
+
+        mean : ndarray of shape (n_features,)
+            Mean of each feature over all samples.
+
+        variance : ndarray of shape (n_features,)
+            Variance of each feature over all samples.
+
+        variation : ndarray of shape (n_features,)
+            Variation of each feature over all samples.
+
+        sum_squares : ndarray of shape (n_features,)
+            Sum of squares for each feature over all samples.
+
+        standard_deviation : ndarray of shape (n_features,)
+            Standard deviation of each feature over all samples.
+
+        sum_squares_centered : ndarray of shape (n_features,)
+            Centered sum of squares for each feature over all samples.
+
+        second_order_raw_moment : ndarray of shape (n_features,)
+            Second order moment of each feature over all samples.
+
+    Notes
+    -----
+        Attributes are populated only for corresponding result options.
+    """
+
+    def __init__(self, result_options="all", algorithm="by_default"):
+        super().__init__(result_options, algorithm)
+        self._reset()
+        self._queue = None
+
+    @bind_default_backend("basic_statistics")
+    def partial_compute_result(self): ...
+
+    @bind_default_backend("basic_statistics")
+    def partial_compute(self, *args, **kwargs): ...
+
+    @bind_default_backend("basic_statistics")
+    def finalize_compute(self, *args, **kwargs): ...
+
+    def _reset(self):
+        self._need_to_finalize = False
+        self._outtype = None
+        self._queue = None
+        # get the _partial_result pointer from backend
+        self._partial_result = self.partial_compute_result()
+
+    def __getstate__(self):
+        # Since finalize_fit can't be dispatched without directly provided queue
+        # and the dispatching policy can't be serialized, the computation is finalized
+        # here and the policy is not saved in serialized data.
+        self.finalize_fit()
+        data = self.__dict__.copy()
+        data.pop("_queue", None)
+
+        return data
+
+    @supports_queue
+    def partial_fit(self, X, sample_weight=None, queue=None):
+        """Generate partial statistics from batch data in `_partial_result`.
+
+        Parameters
+        ----------
+        X : array-like of shape (n_samples, n_features)
+            Training data batch, where `n_samples` is the number of samples
+            in the batch, and `n_features` is the number of features.
+
+        sample_weight : array-like of shape (n_samples,), default=None
+            Individual weights for each sample.
+
+        queue : SyclQueue or None, default=None
+            SYCL Queue object for device code execution. Default
+            value None causes computation on host.
+
+        Returns
+        -------
+        self : object
+            Returns the instance itself.
+        """
+
+        self._queue = queue
+        if not self._outtype:
+            self._outtype = return_type_constructor(X)
+
+        X_table, sample_weight_table = to_table(X, sample_weight, queue=queue)
+
+        if not hasattr(self, "_onedal_params"):
+            self._onedal_params = self._get_onedal_params(False, dtype=X.dtype)
+
+        self._partial_result = self.partial_compute(
+            self._onedal_params, self._partial_result, X_table, sample_weight_table
+        )
+
+        self._need_to_finalize = True
+        self._queue = queue
+
+    def finalize_fit(self):
+        """Finalize statistics from the current `_partial_result`.
+
+        Returns
+        -------
+        self : object
+            Returns the instance itself.
+        """
+        if self._need_to_finalize:
+            with QM.manage_global_queue(self._queue):
+                result = self.finalize_compute(self._onedal_params, self._partial_result)
+
+            for opt in self.options:
+                setattr(
+                    self,
+                    opt + "_",
+                    from_table(getattr(result, opt), like=self._outtype)[0, :],
+                )
+
+            self._outtype = None
+            self._need_to_finalize = False
+
+        return self

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/onedal/basic_statistics/tests/test_basic_statistics.py

@@ -0,0 +1,241 @@
+# ==============================================================================
+# 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.
+# ==============================================================================
+
+import numpy as np
+import pytest
+from numpy.testing import assert_allclose
+from scipy import sparse as sp
+
+from onedal.basic_statistics import BasicStatistics
+from onedal.basic_statistics.tests.utils import options_and_tests
+from onedal.tests.utils._device_selection import get_queues
+
+options_and_tests_csr = [
+    ("sum", "sum", (5e-6, 1e-9)),
+    ("min", "min", (0, 0)),
+    ("max", "max", (0, 0)),
+    ("mean", "mean", (5e-6, 1e-9)),
+]
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("result_option", options_and_tests.keys())
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_single_option_on_random_data(
+    queue, result_option, row_count, column_count, weighted, dtype
+):
+    function, tols = options_and_tests[result_option]
+    fp32tol, fp64tol = tols
+    seed = 77
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+    else:
+        weights = None
+
+    basicstat = BasicStatistics(result_options=result_option)
+
+    result = basicstat.fit(data, sample_weight=weights, queue=queue)
+
+    res = getattr(result, result_option + "_")
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+        gtr = function(weighted_data)
+    else:
+        gtr = function(data)
+
+    tol = fp32tol if res.dtype == np.float32 else fp64tol
+    assert_allclose(gtr, res, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_multiple_options_on_random_data(queue, row_count, column_count, weighted, dtype):
+    seed = 42
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+    else:
+        weights = None
+
+    basicstat = BasicStatistics(result_options=["mean", "max", "sum"])
+
+    result = basicstat.fit(data, sample_weight=weights, queue=queue)
+
+    res_mean, res_max, res_sum = result.mean_, result.max_, result.sum_
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+        gtr_mean, gtr_max, gtr_sum = (
+            options_and_tests["mean"][0](weighted_data),
+            options_and_tests["max"][0](weighted_data),
+            options_and_tests["sum"][0](weighted_data),
+        )
+    else:
+        gtr_mean, gtr_max, gtr_sum = (
+            options_and_tests["mean"][0](data),
+            options_and_tests["max"][0](data),
+            options_and_tests["sum"][0](data),
+        )
+
+    tol = 5e-4 if res_mean.dtype == np.float32 else 1e-7
+    assert_allclose(gtr_mean, res_mean, atol=tol)
+    assert_allclose(gtr_max, res_max, atol=tol)
+    assert_allclose(gtr_sum, res_sum, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_all_option_on_random_data(queue, row_count, column_count, weighted, dtype):
+    seed = 77
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+    else:
+        weights = None
+
+    basicstat = BasicStatistics(result_options="all")
+
+    result = basicstat.fit(data, sample_weight=weights, queue=queue)
+
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+
+    for result_option in options_and_tests:
+        function, tols = options_and_tests[result_option]
+        fp32tol, fp64tol = tols
+        res = getattr(result, result_option + "_")
+        if weighted:
+            gtr = function(weighted_data)
+        else:
+            gtr = function(data)
+        tol = fp32tol if res.dtype == np.float32 else fp64tol
+        assert_allclose(gtr, res, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("result_option", options_and_tests.keys())
+@pytest.mark.parametrize("data_size", [100, 1000])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_1d_input_on_random_data(queue, result_option, data_size, weighted, dtype):
+
+    function, tols = options_and_tests[result_option]
+    fp32tol, fp64tol = tols
+    seed = 77
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=data_size)
+    data = data.astype(dtype=dtype)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=data_size)
+        weights = weights.astype(dtype=dtype)
+    else:
+        weights = None
+
+    basicstat = BasicStatistics(result_options=result_option)
+
+    result = basicstat.fit(data, sample_weight=weights, queue=queue)
+
+    res = getattr(result, result_option + "_")
+    if weighted:
+        weighted_data = weights * data
+        gtr = function(weighted_data)
+    else:
+        gtr = function(data)
+
+    tol = fp32tol if res.dtype == np.float32 else fp64tol
+    assert_allclose(gtr, res, atol=tol)
+
+
+@pytest.mark.skipif(not hasattr(sp, "random_array"), reason="requires scipy>=1.12.0")
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_basic_csr(queue, dtype):
+    seed = 42
+    row_count, column_count = 5000, 3008
+
+    gen = np.random.default_rng(seed)
+
+    data = sp.random_array(
+        shape=(row_count, column_count),
+        density=0.01,
+        format="csr",
+        dtype=dtype,
+        random_state=gen,
+    )
+
+    basicstat = BasicStatistics(result_options="mean")
+    result = basicstat.fit(data, queue=queue)
+
+    res_mean = result.mean_
+    gtr_mean = data.mean(axis=0)
+    tol = 5e-6 if res_mean.dtype == np.float32 else 1e-9
+    assert_allclose(gtr_mean, res_mean, rtol=tol)
+
+
+@pytest.mark.skipif(not hasattr(sp, "random_array"), reason="requires scipy>=1.12.0")
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("option", options_and_tests_csr)
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_options_csr(queue, option, dtype):
+    result_option, function, tols = option
+    fp32tol, fp64tol = tols
+
+    if result_option == "max":
+        pytest.skip("There is a bug in oneDAL's max computations on GPU")
+
+    seed = 42
+    row_count, column_count = 20046, 4007
+
+    gen = np.random.default_rng(seed)
+
+    data = sp.random_array(
+        shape=(row_count, column_count),
+        density=0.002,
+        format="csr",
+        dtype=dtype,
+        random_state=gen,
+    )
+
+    basicstat = BasicStatistics(result_options=result_option)
+    result = basicstat.fit(data, queue=queue)
+
+    res = getattr(result, result_option + "_")
+    func = getattr(data, function)
+    gtr = func(axis=0)
+    if type(gtr).__name__ != "ndarray":
+        gtr = gtr.toarray().flatten()
+    tol = fp32tol if res.dtype == np.float32 else fp64tol
+
+    assert_allclose(gtr, res, rtol=tol)

scikit_learn_intelex-2025.10.0.data/data/Lib/site-packages/onedal/basic_statistics/tests/test_incremental_basic_statistics.py

@@ -0,0 +1,279 @@
+# ==============================================================================
+# 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 import IncrementalBasicStatistics
+from onedal.basic_statistics.tests.utils import options_and_tests
+from onedal.datatypes import from_table
+from onedal.tests.utils._device_selection import get_queues
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_multiple_options_on_gold_data(queue, weighted, dtype):
+    X = np.array([[0, 0], [1, 1]])
+    X = X.astype(dtype=dtype)
+    X_split = np.array_split(X, 2)
+    if weighted:
+        weights = np.array([1, 0.5])
+        weights = weights.astype(dtype=dtype)
+        weights_split = np.array_split(weights, 2)
+
+    incbs = IncrementalBasicStatistics()
+    for i in range(2):
+        if weighted:
+            incbs.partial_fit(X_split[i], weights_split[i], queue=queue)
+        else:
+            incbs.partial_fit(X_split[i], queue=queue)
+
+    result = incbs.finalize_fit()
+
+    if weighted:
+        expected_weighted_mean = np.array([0.25, 0.25])
+        expected_weighted_min = np.array([0, 0])
+        expected_weighted_max = np.array([0.5, 0.5])
+        assert_allclose(expected_weighted_mean, result.mean_)
+        assert_allclose(expected_weighted_max, result.max_)
+        assert_allclose(expected_weighted_min, result.min_)
+    else:
+        expected_mean = np.array([0.5, 0.5])
+        expected_min = np.array([0, 0])
+        expected_max = np.array([1, 1])
+        assert_allclose(expected_mean, result.mean_)
+        assert_allclose(expected_max, result.max_)
+        assert_allclose(expected_min, result.min_)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("num_batches", [2, 10])
+@pytest.mark.parametrize("result_option", options_and_tests.keys())
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_single_option_on_random_data(
+    queue, num_batches, result_option, row_count, column_count, weighted, dtype
+):
+    function, tols = options_and_tests[result_option]
+    fp32tol, fp64tol = tols
+    seed = 77
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+    data_split = np.array_split(data, num_batches)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+        weights_split = np.array_split(weights, num_batches)
+    incbs = IncrementalBasicStatistics(result_options=result_option)
+
+    for i in range(num_batches):
+        if weighted:
+            incbs.partial_fit(data_split[i], weights_split[i], queue=queue)
+        else:
+            incbs.partial_fit(data_split[i], queue=queue)
+    result = incbs.finalize_fit()
+
+    res = getattr(result, result_option + "_")
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+        gtr = function(weighted_data)
+    else:
+        gtr = function(data)
+
+    tol = fp32tol if res.dtype == np.float32 else fp64tol
+    assert_allclose(gtr, res, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("num_batches", [2, 10])
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_multiple_options_on_random_data(
+    queue, num_batches, row_count, column_count, weighted, dtype
+):
+    seed = 42
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+    data_split = np.array_split(data, num_batches)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+        weights_split = np.array_split(weights, num_batches)
+    incbs = IncrementalBasicStatistics(result_options=["mean", "max", "sum"])
+
+    for i in range(num_batches):
+        if weighted:
+            incbs.partial_fit(data_split[i], weights_split[i], queue=queue)
+        else:
+            incbs.partial_fit(data_split[i], queue=queue)
+    result = incbs.finalize_fit()
+
+    res_mean, res_max, res_sum = result.mean_, result.max_, result.sum_
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+        gtr_mean, gtr_max, gtr_sum = (
+            options_and_tests["mean"][0](weighted_data),
+            options_and_tests["max"][0](weighted_data),
+            options_and_tests["sum"][0](weighted_data),
+        )
+    else:
+        gtr_mean, gtr_max, gtr_sum = (
+            options_and_tests["mean"][0](data),
+            options_and_tests["max"][0](data),
+            options_and_tests["sum"][0](data),
+        )
+
+    tol = 3e-4 if res_mean.dtype == np.float32 else 1e-7
+    assert_allclose(gtr_mean, res_mean, atol=tol)
+    assert_allclose(gtr_max, res_max, atol=tol)
+    assert_allclose(gtr_sum, res_sum, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("num_batches", [2, 10])
+@pytest.mark.parametrize("row_count", [100, 1000])
+@pytest.mark.parametrize("column_count", [10, 100])
+@pytest.mark.parametrize("weighted", [True, False])
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_all_option_on_random_data(
+    queue, num_batches, row_count, column_count, weighted, dtype
+):
+    seed = 77
+    gen = np.random.default_rng(seed)
+    data = gen.uniform(low=-0.3, high=+0.7, size=(row_count, column_count))
+    data = data.astype(dtype=dtype)
+    data_split = np.array_split(data, num_batches)
+    if weighted:
+        weights = gen.uniform(low=-0.5, high=+1.0, size=row_count)
+        weights = weights.astype(dtype=dtype)
+        weights_split = np.array_split(weights, num_batches)
+    incbs = IncrementalBasicStatistics(result_options="all")
+
+    for i in range(num_batches):
+        if weighted:
+            incbs.partial_fit(data_split[i], weights_split[i], queue=queue)
+        else:
+            incbs.partial_fit(data_split[i], queue=queue)
+    result = incbs.finalize_fit()
+
+    if weighted:
+        weighted_data = np.diag(weights) @ data
+
+    for result_option in options_and_tests:
+        function, tols = options_and_tests[result_option]
+        fp32tol, fp64tol = tols
+        res = getattr(result, result_option + "_")
+        if weighted:
+            gtr = function(weighted_data)
+        else:
+            gtr = function(data)
+        tol = fp32tol if res.dtype == np.float32 else fp64tol
+        assert_allclose(gtr, res, atol=tol)
+
+
+@pytest.mark.parametrize("queue", get_queues())
+@pytest.mark.parametrize("dtype", [np.float32, np.float64])
+def test_incremental_estimator_pickle(queue, dtype):
+    import pickle
+
+    from onedal.basic_statistics import IncrementalBasicStatistics
+
+    incbs = IncrementalBasicStatistics()
+
+    # Check that estimator can be serialized without any data.
+    dump = pickle.dumps(incbs)
+    incbs_loaded = pickle.loads(dump)
+    seed = 77
+    gen = np.random.default_rng(seed)
+    X = gen.uniform(low=-0.3, high=+0.7, size=(10, 10))
+    X = X.astype(dtype)
+    X_split = np.array_split(X, 2)
+    incbs.partial_fit(X_split[0], queue=queue)
+    incbs_loaded.partial_fit(X_split[0], queue=queue)
+
+    assert incbs._need_to_finalize == True
+    assert incbs_loaded._need_to_finalize == True
+
+    # Check that estimator can be serialized after partial_fit call.
+    dump = pickle.dumps(incbs)
+    incbs_loaded = pickle.loads(dump)
+    assert incbs._need_to_finalize == False
+    # Finalize is called during serialization to make sure partial results are finalized correctly.
+    assert incbs_loaded._need_to_finalize == False
+
+    partial_n_rows = from_table(incbs._partial_result.partial_n_rows)
+    partial_n_rows_loaded = from_table(incbs_loaded._partial_result.partial_n_rows)
+    assert_allclose(partial_n_rows, partial_n_rows_loaded)
+
+    partial_min = from_table(incbs._partial_result.partial_min)
+    partial_min_loaded = from_table(incbs_loaded._partial_result.partial_min)
+    assert_allclose(partial_min, partial_min_loaded)
+
+    partial_max = from_table(incbs._partial_result.partial_max)
+    partial_max_loaded = from_table(incbs_loaded._partial_result.partial_max)
+    assert_allclose(partial_max, partial_max_loaded)
+
+    partial_sum = from_table(incbs._partial_result.partial_sum)
+    partial_sum_loaded = from_table(incbs_loaded._partial_result.partial_sum)
+    assert_allclose(partial_sum, partial_sum_loaded)
+
+    partial_sum_squares = from_table(incbs._partial_result.partial_sum_squares)
+    partial_sum_squares_loaded = from_table(
+        incbs_loaded._partial_result.partial_sum_squares
+    )
+    assert_allclose(partial_sum_squares, partial_sum_squares_loaded)
+
+    partial_sum_squares_centered = from_table(
+        incbs._partial_result.partial_sum_squares_centered
+    )
+    partial_sum_squares_centered_loaded = from_table(
+        incbs_loaded._partial_result.partial_sum_squares_centered
+    )
+    assert_allclose(partial_sum_squares_centered, partial_sum_squares_centered_loaded)
+
+    incbs.partial_fit(X_split[1], queue=queue)
+    incbs_loaded.partial_fit(X_split[1], queue=queue)
+    assert incbs._need_to_finalize == True
+    assert incbs_loaded._need_to_finalize == True
+
+    dump = pickle.dumps(incbs_loaded)
+    incbs_loaded = pickle.loads(dump)
+
+    assert incbs._need_to_finalize == True
+    assert incbs_loaded._need_to_finalize == False
+
+    incbs.finalize_fit()
+    incbs_loaded.finalize_fit()
+
+    # Check that finalized estimator can be serialized.
+    dump = pickle.dumps(incbs_loaded)
+    incbs_loaded = pickle.loads(dump)
+
+    for result_option in options_and_tests:
+        _, tols = options_and_tests[result_option]
+        fp32tol, fp64tol = tols
+        res = getattr(incbs, result_option + "_")
+        res_loaded = getattr(incbs_loaded, result_option + "_")
+        tol = fp32tol if res.dtype == np.float32 else fp64tol
+        assert_allclose(res, res_loaded, atol=tol)