pytme 0.2.3__cp311-cp311-macosx_14_0_arm64.whl → 0.2.4__cp311-cp311-macosx_14_0_arm64.whl

tests/test_density.py

@@ -0,0 +1,508 @@
+from os import remove
+from tempfile import mkstemp
+from itertools import permutations
+from importlib_resources import files
+
+import pytest
+import numpy as np
+
+from tme import Density, Structure, Preprocessor
+from tme.matching_utils import create_mask, euler_to_rotationmatrix
+
+DEFAULT_DATA = create_mask(
+    mask_type="ellipse",
+    center=(20, 20, 20),
+    radius=(10, 5, 10),
+    shape=(50, 50, 50),
+)
+DEFAULT_DATA = Preprocessor().gaussian_filter(DEFAULT_DATA * 10, sigma=2)
+DEFAULT_DATA = DEFAULT_DATA.astype(np.float32)
+DEFAULT_ORIGIN = np.array([0, 0, 0])
+DEFAULT_SAMPLING_RATE = np.array([1, 1, 1])
+
+BASEPATH = files("tests.data")
+
+
+class TestDensity:
+    def setup_method(self):
+        self.density = Density(
+            data=DEFAULT_DATA,
+            origin=DEFAULT_ORIGIN,
+            sampling_rate=DEFAULT_SAMPLING_RATE,
+            metadata={
+                "min": DEFAULT_DATA.min(),
+                "max": DEFAULT_DATA.max(),
+                "mean": DEFAULT_DATA.mean(),
+                "std": DEFAULT_DATA.std(),
+            },
+        )
+        _, self.path = mkstemp()
+        self.structure_path = str(BASEPATH.joinpath("Structures/5khe.cif"))
+
+    def teardown_method(self):
+        del self.density
+        remove(self.path)
+
+    def test_initialization(self):
+        data = DEFAULT_DATA
+        origin = DEFAULT_ORIGIN
+        sampling_rate = DEFAULT_SAMPLING_RATE
+        metadata = {"test_key": "test_value"}
+
+        density = Density(data, origin, sampling_rate, metadata)
+
+        assert np.array_equal(density.data, data)
+        assert np.array_equal(density.origin, origin)
+        assert np.array_equal(density.sampling_rate, sampling_rate)
+        assert density.metadata == metadata
+
+    @pytest.mark.parametrize(
+        "data,origin,sampling_rate,metadata",
+        [
+            (np.random.rand(50, 50, 50), (0, 0, 0), (1, 2), {}),
+            (np.random.rand(50, 50, 50), (0, 0, 0), (1, 2, 3, 4), {}),
+            (np.random.rand(50, 50, 50), (0, 0, 0), (1, 2, 3), "not_a_dict"),
+            (np.random.rand(50, 50, 50), (0, 0), (1, 2, 3), "not_a_dict"),
+        ],
+    )
+    def test_initialization_errors(self, data, origin, sampling_rate, metadata):
+        with pytest.raises(ValueError):
+            Density(data, origin, sampling_rate, metadata)
+
+    def test_repr(self):
+        data = DEFAULT_DATA
+        origin = DEFAULT_ORIGIN
+        sampling_rate = DEFAULT_SAMPLING_RATE
+
+        density = Density(data, origin, sampling_rate)
+        repr_str = density.__repr__()
+
+        response = "Density object at {}\nOrigin: {}, Sampling Rate: {}, Shape: {}"
+        response = response.format(
+            hex(id(density)),
+            tuple(round(float(x), 3) for x in density.origin),
+            tuple(round(float(x), 3) for x in density.sampling_rate),
+            density.shape,
+        )
+        assert response == repr_str
+
+    @pytest.mark.parametrize("gzip", [(False), (True)])
+    def test_to_file(self, gzip: bool):
+        self.density.to_file(self.path, gzip=gzip)
+        assert True
+
+    def test_from_file(self):
+        self.test_to_file(gzip=False)
+        density = Density.from_file(self.path)
+        assert np.allclose(density.data, self.density.data)
+        assert np.allclose(density.sampling_rate, self.density.sampling_rate)
+        assert np.allclose(density.origin, self.density.origin)
+        assert density.metadata == self.density.metadata
+
+    def test_from_file_baseline(self):
+        self.test_to_file(gzip=False)
+        density = Density.from_file(str(BASEPATH.joinpath("Maps/emd_8621.mrc.gz")))
+        assert np.allclose(density.origin, (-1.45, 2.90, 4.35), rtol=0.1)
+        assert np.allclose(density.sampling_rate, (1.45), rtol=0.3)
+
+    @pytest.mark.parametrize("extension", ("mrc", "em", "tiff", "h5"))
+    @pytest.mark.parametrize("gzip", (True, False))
+    @pytest.mark.parametrize("use_memmap", (True, False))
+    @pytest.mark.parametrize("subset", (True, False))
+    def test_file_format_io(self, extension, gzip, subset, use_memmap):
+        base = Density(
+            data=np.random.rand(50, 50, 50), origin=(0, 0, 0), sampling_rate=(1, 1, 1)
+        )
+        data_subset = (slice(0, 22), slice(31, 46), slice(12, 25))
+        if extension not in ("mrc", "em"):
+            base = Density(
+                data=np.random.rand(50, 50), origin=(0, 0), sampling_rate=(1, 1)
+            )
+            data_subset = (slice(0, 22), slice(31, 46))
+        if gzip:
+            use_memmap = False
+
+        if not subset:
+            data_subset = tuple(slice(0, x) for x in base.shape)
+
+        suffix = f".{extension}.gz" if gzip else f".{extension}"
+        _, output_file = mkstemp(suffix=suffix)
+        base.to_file(output_file, gzip=gzip)
+        temp = Density.from_file(output_file, use_memmap=use_memmap, subset=data_subset)
+        assert np.allclose(base.data[data_subset], temp.data)
+        if extension.upper() == "MRC":
+            assert np.allclose(base.origin, temp.origin)
+            assert np.allclose(base.sampling_rate, temp.sampling_rate)
+
+    def test__read_binary_subset_error(self):
+        base = Density(
+            data=np.random.rand(50, 50, 50), origin=(0, 0, 0), sampling_rate=(1, 1, 1)
+        )
+        _, output_file = mkstemp()
+        base.to_file(output_file)
+        with pytest.raises((ValueError, OSError)):
+            Density.from_file(output_file, subset=(slice(0, 10),))
+            Density.from_file(
+                output_file, subset=(slice(-1, 10), slice(5, 10), slice(5, 10))
+            )
+            Density.from_file(
+                output_file, subset=(slice(20, 100), slice(5, 10), slice(5, 10))
+            )
+
+    def test_from_structure(self):
+        _ = Density.from_structure(
+            self.structure_path,
+            origin=self.density.origin,
+            shape=self.density.shape,
+            sampling_rate=self.density.sampling_rate,
+        )
+        _ = Density.from_structure(self.structure_path, shape=(30, 30, 30))
+        _ = Density.from_structure(
+            self.structure_path,
+            shape=(30, 30, 30),
+            origin=self.density.origin,
+        )
+        _ = Density.from_structure(self.structure_path, origin=self.density.origin)
+        _ = Density.from_structure(
+            self.structure_path,
+            shape=(30, 30, 30),
+            sampling_rate=6,
+            origin=self.density.origin,
+        )
+        _ = Density.from_structure(
+            self.structure_path,
+            shape=(30, 30, 30),
+            sampling_rate=None,
+            origin=self.density.origin,
+        )
+        _ = Density.from_structure(
+            self.structure_path, weight_type="atomic_weight", chain="A"
+        )
+
+    @pytest.mark.parametrize(
+        "weight_type",
+        [
+            "atomic_weight",
+            "atomic_number",
+            "van_der_waals_radius",
+            "scattering_factors",
+            "lowpass_scattering_factors",
+            "gaussian",
+        ],
+    )
+    def test_from_structure_weight_types(self, weight_type):
+        _ = Density.from_structure(
+            self.structure_path,
+            weight_type=weight_type,
+        )
+
+    def test_from_structure_weight_types_error(self):
+        with pytest.raises(NotImplementedError):
+            _ = Density.from_structure(
+                self.structure_path,
+                weight_type=None,
+            )
+
+    @pytest.mark.parametrize(
+        "weight_type", ["scattering_factors", "lowpass_scattering_factors"]
+    )
+    @pytest.mark.parametrize(
+        "scattering_factors", ["dt1969", "wk1995", "peng1995", "peng1999"]
+    )
+    def test_from_structure_scattering(self, scattering_factors, weight_type):
+        _ = Density.from_structure(
+            self.structure_path,
+            weight_type=weight_type,
+            weight_type_args={"source": scattering_factors},
+        )
+
+    def test_from_structure_error(self):
+        with pytest.raises(NotImplementedError):
+            _ = Density.from_structure(self.structure_path, weight_type="RAISERROR")
+        with pytest.raises(ValueError):
+            _ = Density.from_structure(self.structure_path, sampling_rate=(1, 5))
+
+    def test_empty(self):
+        empty_density = self.density.empty
+        assert np.allclose(empty_density.data, np.zeros_like(empty_density.data))
+        assert np.allclose(empty_density.sampling_rate, self.density.sampling_rate)
+        assert np.allclose(empty_density.origin, self.density.origin)
+        assert empty_density.metadata == {"min": 0, "max": 0, "mean": 0, "std": 0}
+
+    def test_copy(self):
+        copied_density = self.density.copy()
+        assert np.allclose(copied_density.data, self.density.data)
+        assert np.allclose(copied_density.sampling_rate, self.density.sampling_rate)
+        assert np.allclose(copied_density.origin, self.density.origin)
+        assert copied_density.metadata == self.density.metadata
+
+    def test_to_memmap(self):
+        filename = self.path
+
+        temp = self.density.copy()
+        shape, dtype = temp.data.shape, temp.data.dtype
+
+        arr_memmap = np.memmap(filename, mode="w+", dtype=dtype, shape=shape)
+        arr_memmap[:] = temp.data[:]
+        arr_memmap.flush()
+        arr_memmap = np.memmap(filename, mode="r", dtype=dtype, shape=shape)
+
+        temp.to_memmap()
+
+        assert np.allclose(temp.data, arr_memmap)
+
+    def test_to_numpy(self):
+        temp = self.density.copy()
+        temp.to_memmap()
+        temp.to_numpy()
+
+        assert np.allclose(temp.data, self.density.data)
+
+    @pytest.mark.parametrize("threshold", [(0), (0.5), (1), (1.5)])
+    def test_to_pointcloud(self, threshold):
+        indices = self.density.to_pointcloud(threshold=threshold)
+        assert indices.shape[0] == self.density.data.ndim
+        assert np.all(self.density.data[tuple(indices)] > threshold)
+
+    def test__pad_slice(self):
+        x, y, z = self.density.shape
+        box = (slice(-5, z + 5), slice(0, y - 5), slice(2, z + 2))
+        padded_data = self.density._pad_slice(box)
+        assert padded_data.shape == (60, y, z)
+
+    def test_adjust_box(self):
+        box = (slice(10, 40), slice(10, 40), slice(10, 40))
+        self.density.adjust_box(box)
+        assert self.density.data.shape == (30, 30, 30)
+        np.testing.assert_array_equal(
+            self.density.origin, DEFAULT_ORIGIN + 10 * DEFAULT_SAMPLING_RATE
+        )
+
+    def test_trim_box(self):
+        toy_data = np.zeros((20, 20, 20))
+        signal = np.ones((5, 5, 5))
+        toy_data[0:5, 5:10, 10:15] = signal
+        temp = self.density.empty
+        temp.data = toy_data
+        trim_box = temp.trim_box(cutoff=0.5, margin=0)
+        trim_box_margin = temp.trim_box(cutoff=0.5, margin=2)
+        temp.adjust_box(trim_box)
+        assert np.allclose(temp.data, signal)
+        assert trim_box_margin == tuple((slice(0, 7), slice(3, 12), slice(8, 17)))
+
+    def test_pad(self):
+        new_shape = (70, 70, 70)
+        self.density.pad(new_shape)
+        assert self.density.data.shape == new_shape
+
+    @pytest.mark.parametrize("new_shape", [(70,), (70, 70), (70, 70, 70, 70)])
+    def test_pad_error(self, new_shape):
+        with pytest.raises(ValueError):
+            self.density.pad(new_shape)
+
+    def test_minimum_enclosing_box(self):
+        # The exact shape may vary, so we will mainly ensure that
+        # the data is correctly adapted.
+        # Further, more precise tests could be added.
+        temp = self.density.copy()
+        box = temp.minimum_enclosing_box(cutoff=0.5)
+        assert len(box) == temp.data.ndim
+        temp = self.density.copy()
+        box = temp.minimum_enclosing_box(cutoff=0.5, use_geometric_center=True)
+        assert len(box) == temp.data.ndim
+
+    @pytest.mark.parametrize(
+        "cutoff", [DEFAULT_DATA.min() - 1, 0, DEFAULT_DATA.max() - 0.1]
+    )
+    def test_centered(self, cutoff):
+        centered_density, translation = self.density.centered(cutoff=cutoff)
+        com = centered_density.center_of_mass(centered_density.data, 0)
+
+        difference = np.abs(
+            np.subtract(
+                np.rint(np.array(com)).astype(int),
+                np.array(centered_density.shape) // 2,
+            )
+        )
+        assert np.all(difference <= self.density.sampling_rate)
+
+    @pytest.mark.parametrize("use_geometric_center", (True, False))
+    @pytest.mark.parametrize("order", (1, 3))
+    def test_rigid_transform(self, use_geometric_center: bool, order: int):
+        temp = self.density.copy()
+        if use_geometric_center:
+            box = temp.minimum_enclosing_box(cutoff=0, use_geometric_center=True)
+            temp.adjust_box(box)
+        else:
+            temp, translation = temp.centered()
+
+        swaps = set(permutations([0, 1, 2]))
+        temp_matrix = np.eye(temp.data.ndim).astype(np.float32)
+        rotation_matrix = np.zeros_like(temp_matrix)
+
+        initial_weight = np.sum(np.abs(temp.data))
+        for z, y, x in swaps:
+            rotation_matrix[:, 0] = temp_matrix[:, z]
+            rotation_matrix[:, 1] = temp_matrix[:, y]
+            rotation_matrix[:, 2] = temp_matrix[:, x]
+
+            transformed = temp.rigid_transform(
+                rotation_matrix=rotation_matrix,
+                translation=np.zeros(temp.data.ndim),
+                use_geometric_center=use_geometric_center,
+                order=order,
+            )
+            transformed_weight = np.sum(np.abs(transformed.data))
+            assert np.abs(1 - initial_weight / transformed_weight) < 0.01
+
+    @pytest.mark.parametrize(
+        "new_sampling_rate,order",
+        [(2, 1), (4, 3)],
+    )
+    @pytest.mark.parametrize("method", ("spline", "fourier"))
+    def test_resample(self, new_sampling_rate, order, method):
+        resampled = self.density.resample(
+            new_sampling_rate=new_sampling_rate, order=order, method=method
+        )
+        assert np.allclose(
+            resampled.shape,
+            np.divide(self.density.shape, new_sampling_rate).astype(int),
+        )
+
+    @pytest.mark.parametrize(
+        "fraction_surface,volume_factor",
+        [(0.5, 2), (1, 3)],
+    )
+    def test_density_boundary(self, fraction_surface, volume_factor):
+        # TODO: Pre compute volume boundary on real data
+        boundary = self.density.density_boundary(
+            weight=1000, fraction_surface=fraction_surface, volume_factor=volume_factor
+        )
+        assert boundary[0] < boundary[1]
+
+    @pytest.mark.parametrize(
+        "fraction_surface,volume_factor",
+        [(-0.5, 0), (1, -3)],
+    )
+    def test_density_boundary_error(self, fraction_surface, volume_factor):
+        with pytest.raises(ValueError):
+            _ = self.density.density_boundary(
+                weight=1000,
+                fraction_surface=fraction_surface,
+                volume_factor=volume_factor,
+            )
+
+    @pytest.mark.parametrize(
+        "method",
+        [("ConvexHull"), ("Weight"), ("Sobel"), ("Laplace"), ("Minimum")],
+    )
+    def test_surface_coordinates(self, method):
+        density_boundaries = self.density.density_boundary(weight=1000)
+        self.density.surface_coordinates(
+            density_boundaries=density_boundaries, method=method
+        )
+
+    def test_surface_coordinates_error(self):
+        density_boundaries = self.density.density_boundary(weight=1000)
+        with pytest.raises(ValueError):
+            self.density.surface_coordinates(
+                density_boundaries=density_boundaries, method=None
+            )
+
+    def test_normal_vectors(self):
+        density_boundaries = self.density.density_boundary(weight=1000)
+        coordinates = self.density.surface_coordinates(
+            density_boundaries=density_boundaries, method="ConvexHull"
+        )
+        self.density.normal_vectors(coordinates=coordinates)
+
+    def test_normal_vectors_error(self):
+        coordinates = np.random.rand(10, 10, 10)
+        with pytest.raises(ValueError):
+            self.density.normal_vectors(coordinates=coordinates)
+
+        coordinates = np.random.rand(10, 4)
+        with pytest.raises(ValueError):
+            self.density.normal_vectors(coordinates=coordinates)
+
+    def test_core_mask(self):
+        mask = self.density.core_mask()
+        assert mask.sum() > 0
+
+    def test_center_of_mass(self):
+        center, shape, radius = (10, 10), (20, 20), 5
+        n = len(shape)
+        position = np.array(center).reshape((-1,) + (1,) * n)
+        arr = np.linalg.norm(np.indices(shape) - position, axis=0)
+        arr = (arr <= radius).astype(np.float32)
+
+        center_of_mass = Density.center_of_mass(arr)
+        assert np.allclose(center, center_of_mass)
+
+    @pytest.mark.parametrize(
+        "method",
+        [
+            ("CrossCorrelation"),
+            ("NormalizedCrossCorrelation"),
+        ],
+    )
+    def test_match_densities(self, method: str):
+        target = np.zeros((30, 30, 30))
+        target[5:10, 15:22, 10:13] = 1
+
+        target = Density(target, sampling_rate=(1, 1, 1), origin=(0, 0, 0))
+        target, translation = target.centered(cutoff=0)
+
+        template = target.copy()
+
+        initial_translation = np.array([-1, 3, 0])
+        initial_rotation = euler_to_rotationmatrix((-10, 2, 5))
+
+        template = template.rigid_transform(
+            rotation_matrix=initial_rotation,
+            translation=initial_translation,
+            use_geometric_center=False,
+        )
+
+        target.sampling_rate = np.array(target.sampling_rate[0])
+        template.sampling_rate = np.array(template.sampling_rate[0])
+
+        aligned, translation, rotation = Density.match_densities(
+            target=target, template=template, scoring_method=method, maxiter=5
+        )
+        assert np.allclose(-translation, initial_translation, atol=2)
+        assert np.allclose(np.linalg.inv(rotation), initial_rotation, atol=0.2)
+
+    def test_match_structure_to_density(self):
+        density = Density.from_file("tests/data/Maps/emd_8621.mrc.gz")
+        density = density.resample(density.sampling_rate * 4)
+        structure = Structure.from_file(
+            "tests/data/Structures/5uz4.cif", filter_by_residues=None
+        )
+
+        initial_translation = np.array([-1, 0, 5])
+        initial_rotation = euler_to_rotationmatrix((-10, 2, 5))
+        structure.rigid_transform(
+            translation=initial_translation, rotation_matrix=initial_rotation
+        )
+        np.random.seed(12)
+        ret = Density.match_structure_to_density(
+            target=density,
+            template=structure,
+            cutoff_target=0,
+            scoring_method="CrossCorrelation",
+            maxiter=10,
+        )
+        structure_aligned, translation, rotation_matrix = ret
+
+        assert np.allclose(
+            structure_aligned.atom_coordinate.shape, structure.atom_coordinate.shape
+        )
+        assert np.allclose(-translation, initial_translation, atol=2)
+        assert np.allclose(np.linalg.inv(rotation_matrix), initial_rotation, atol=0.2)
+
+    def test_fourier_shell_correlation(self):
+        fsc = Density.fourier_shell_correlation(
+            self.density.copy(), self.density.copy()
+        )
+        assert fsc.shape[1] == 2

tests/test_extensions.py

@@ -0,0 +1,130 @@
+import pytest
+import numpy as np
+
+from scipy.spatial import distance
+
+from tme.extensions import (
+    absolute_minimum_deviation,
+    max_euclidean_distance,
+    find_candidate_indices,
+    find_candidate_coordinates,
+    max_index_by_label,
+    online_statistics,
+)
+
+
+COORDINATES, N_COORDINATES = {}, 50
+for i in range(1, 4):
+    COORDINATES[i] = np.random.choice(np.arange(100), size=50 * i).reshape(
+        N_COORDINATES, i
+    )
+
+np.random.seed(42)
+TEST_DATA = 10 * np.random.rand(5000)
+
+
+class TestExtensions:
+    @pytest.mark.parametrize("dimension", list(COORDINATES.keys()))
+    @pytest.mark.parametrize("dtype", [np.int32, int, np.float32, np.float64])
+    def test_absolute_minimum_deviation(self, dimension, dtype):
+        coordinates = COORDINATES[dimension].astype(dtype)
+        output = np.zeros(
+            (coordinates.shape[0], coordinates.shape[0]), dtype=coordinates.dtype
+        )
+        absolute_minimum_deviation(coordinates=coordinates, output=output)
+        expected_output = distance.cdist(
+            coordinates, coordinates, lambda u, v: np.min(np.abs(u - v))
+        )
+        assert np.allclose(output, expected_output)
+
+    @pytest.mark.parametrize("dimension", list(COORDINATES.keys()))
+    @pytest.mark.parametrize("dtype", [np.int32, int, np.float32, np.float64])
+    def test_max_euclidean_distance(self, dimension, dtype):
+        coordinates = COORDINATES[dimension].astype(dtype)
+        print(coordinates.shape)
+
+        max_distance, pair = max_euclidean_distance(coordinates=coordinates)
+        distances = distance.cdist(coordinates, coordinates, "euclidean")
+        distances_max = distances.max()
+        assert np.allclose(max_distance, distances_max)
+
+    @pytest.mark.parametrize("dimension", list(COORDINATES.keys()))
+    @pytest.mark.parametrize("dtype", [np.int32, int, np.float32, np.float64])
+    @pytest.mark.parametrize("min_distance", [0, 5, 10])
+    def test_find_candidate_indices(self, dimension, dtype, min_distance):
+        coordinates = COORDINATES[dimension].astype(dtype)
+        print(coordinates.shape)
+
+        min_distance = np.array([min_distance]).astype(dtype)[0]
+
+        candidates = find_candidate_indices(
+            coordinates=coordinates, min_distance=min_distance
+        )
+
+        distances = distance.cdist(coordinates[candidates], coordinates[candidates])
+        np.fill_diagonal(distances, np.inf)
+        assert np.all(distances >= min_distance)
+
+    @pytest.mark.parametrize("dimension", list(COORDINATES.keys()))
+    @pytest.mark.parametrize("dtype", [np.int32, int, np.float32, np.float64])
+    @pytest.mark.parametrize("min_distance", [0, 5, 10])
+    def test_find_candidate_coordinates(self, dimension, dtype, min_distance):
+        coordinates = COORDINATES[dimension].astype(dtype)
+        print(coordinates.shape)
+
+        min_distance = np.array([min_distance]).astype(dtype)[0]
+
+        filtered_coordinates = find_candidate_coordinates(
+            coordinates=coordinates, min_distance=min_distance
+        )
+
+        distances = distance.cdist(filtered_coordinates, filtered_coordinates)
+        np.fill_diagonal(distances, np.inf)
+        assert np.all(distances >= min_distance)
+
+    @pytest.mark.parametrize("dtype_labels", [np.int32, int, np.float32, np.float64])
+    @pytest.mark.parametrize("dtype_scores", [np.int32, int, np.float32, np.float64])
+    def test_max_index_by_label(self, dtype_labels, dtype_scores):
+        labels = np.array([1, 1, 2, 3, 3, 3, 2], dtype=dtype_labels)
+        scores = np.array([0.5, 0.8, 0.7, 0.2, 0.9, 0.6, 0.5])
+        scores = (10 * scores).astype(dtype_scores)
+        expected_result = {}
+        for label in np.unique(labels):
+            mask = labels == label
+            expected_result[label] = np.argmax(scores * mask)
+
+        ret = max_index_by_label(labels=labels, scores=scores)
+        print(ret)
+        for k in expected_result:
+            assert np.allclose(expected_result[k], ret[k])
+
+    @pytest.mark.parametrize("dtype", [np.int32, int, np.float32, np.float64])
+    @pytest.mark.parametrize("splits", [1, 5, 10])
+    def test_online_statistics(self, dtype, splits):
+        parts = np.array_split(TEST_DATA, splits)
+        n, rmean, ssqd, reference = 0, 0, 0, 0.0
+        better_or_equal = 0
+        end_idx = 0
+        for part in parts:
+            start_idx = end_idx
+            end_idx = start_idx + len(part)
+
+            n, rmean, ssqd, nbetter_or_equal, max_value, min_value = online_statistics(
+                arr=part, n=n, rmean=rmean, ssqd=ssqd
+            )
+            better_or_equal += nbetter_or_equal
+
+            data = TEST_DATA[:end_idx]
+            dn = len(data)
+            drmean = np.mean(data)
+            dssqd = np.sum((data - rmean) ** 2)
+            dnbetter_or_equal = np.sum(data >= reference)
+            dmax_value = np.max(data)
+            dmin_value = np.min(data)
+
+            assert n == dn
+            assert np.isclose(rmean, drmean, atol=1e-8)
+            assert np.isclose(ssqd, dssqd, atol=1e-8)
+            assert better_or_equal == dnbetter_or_equal
+            assert np.isclose(max_value, dmax_value, atol=1e-1)
+            assert np.isclose(min_value, dmin_value, atol=1e-1)