scipy 1.16.2__cp312-cp312-win_arm64.whl

scipy/sparse/csgraph/tests/test_graph_laplacian.py

@@ -0,0 +1,368 @@
+import pytest
+import numpy as np
+from numpy.testing import assert_allclose
+from pytest import raises as assert_raises
+from scipy import sparse
+
+from scipy.sparse import csgraph
+from scipy._lib._util import np_long, np_ulong
+
+
+def check_int_type(mat):
+    return np.issubdtype(mat.dtype, np.signedinteger) or np.issubdtype(
+        mat.dtype, np_ulong
+    )
+
+
+def test_laplacian_value_error():
+    for t in int, float, complex:
+        for m in ([1, 1],
+                  [[[1]]],
+                  [[1, 2, 3], [4, 5, 6]],
+                  [[1, 2], [3, 4], [5, 5]]):
+            A = np.array(m, dtype=t)
+            assert_raises(ValueError, csgraph.laplacian, A)
+
+
+def _explicit_laplacian(x, normed=False):
+    if sparse.issparse(x):
+        x = x.toarray()
+    x = np.asarray(x)
+    y = -1.0 * x
+    for j in range(y.shape[0]):
+        y[j,j] = x[j,j+1:].sum() + x[j,:j].sum()
+    if normed:
+        d = np.diag(y).copy()
+        d[d == 0] = 1.0
+        y /= d[:,None]**.5
+        y /= d[None,:]**.5
+    return y
+
+
+def _check_symmetric_graph_laplacian(mat, normed, copy=True):
+    if not hasattr(mat, 'shape'):
+        mat = eval(mat, dict(np=np, sparse=sparse))
+
+    if sparse.issparse(mat):
+        sp_mat = mat
+        mat = sp_mat.toarray()
+    else:
+        sp_mat = sparse.csr_array(mat)
+
+    mat_copy = np.copy(mat)
+    sp_mat_copy = sparse.csr_array(sp_mat, copy=True)
+
+    n_nodes = mat.shape[0]
+    explicit_laplacian = _explicit_laplacian(mat, normed=normed)
+    laplacian = csgraph.laplacian(mat, normed=normed, copy=copy)
+    sp_laplacian = csgraph.laplacian(sp_mat, normed=normed,
+                                     copy=copy)
+
+    if copy:
+        assert_allclose(mat, mat_copy)
+        _assert_allclose_sparse(sp_mat, sp_mat_copy)
+    else:
+        if not (normed and check_int_type(mat)):
+            assert_allclose(laplacian, mat)
+            if sp_mat.format == 'coo':
+                _assert_allclose_sparse(sp_laplacian, sp_mat)
+
+    assert_allclose(laplacian, sp_laplacian.toarray())
+
+    for tested in [laplacian, sp_laplacian.toarray()]:
+        if not normed:
+            assert_allclose(tested.sum(axis=0), np.zeros(n_nodes))
+        assert_allclose(tested.T, tested)
+        assert_allclose(tested, explicit_laplacian)
+
+
+def test_symmetric_graph_laplacian():
+    symmetric_mats = (
+        'np.arange(10) * np.arange(10)[:, np.newaxis]',
+        'np.ones((7, 7))',
+        'np.eye(19)',
+        'sparse.diags([1, 1], [-1, 1], shape=(4, 4))',
+        'sparse.diags([1, 1], [-1, 1], shape=(4, 4)).toarray()',
+        'sparse.diags([1, 1], [-1, 1], shape=(4, 4)).todense()',
+        'np.vander(np.arange(4)) + np.vander(np.arange(4)).T'
+    )
+    for mat in symmetric_mats:
+        for normed in True, False:
+            for copy in True, False:
+                _check_symmetric_graph_laplacian(mat, normed, copy)
+
+
+def _assert_allclose_sparse(a, b, **kwargs):
+    # helper function that can deal with sparse matrices
+    if sparse.issparse(a):
+        a = a.toarray()
+    if sparse.issparse(b):
+        b = b.toarray()
+    assert_allclose(a, b, **kwargs)
+
+
+def _check_laplacian_dtype_none(
+    A, desired_L, desired_d, normed, use_out_degree, copy, dtype, arr_type
+):
+    mat = arr_type(A, dtype=dtype)
+    L, d = csgraph.laplacian(
+        mat,
+        normed=normed,
+        return_diag=True,
+        use_out_degree=use_out_degree,
+        copy=copy,
+        dtype=None,
+    )
+    if normed and check_int_type(mat):
+        assert L.dtype == np.float64
+        assert d.dtype == np.float64
+        _assert_allclose_sparse(L, desired_L, atol=1e-12)
+        _assert_allclose_sparse(d, desired_d, atol=1e-12)
+    else:
+        assert L.dtype == dtype
+        assert d.dtype == dtype
+        desired_L = np.asarray(desired_L).astype(dtype)
+        desired_d = np.asarray(desired_d).astype(dtype)
+        _assert_allclose_sparse(L, desired_L, atol=1e-12)
+        _assert_allclose_sparse(d, desired_d, atol=1e-12)
+
+    if not copy:
+        if not (normed and check_int_type(mat)):
+            if type(mat) is np.ndarray:
+                assert_allclose(L, mat)
+            elif mat.format == "coo":
+                _assert_allclose_sparse(L, mat)
+
+
+def _check_laplacian_dtype(
+    A, desired_L, desired_d, normed, use_out_degree, copy, dtype, arr_type
+):
+    mat = arr_type(A, dtype=dtype)
+    L, d = csgraph.laplacian(
+        mat,
+        normed=normed,
+        return_diag=True,
+        use_out_degree=use_out_degree,
+        copy=copy,
+        dtype=dtype,
+    )
+    assert L.dtype == dtype
+    assert d.dtype == dtype
+    desired_L = np.asarray(desired_L).astype(dtype)
+    desired_d = np.asarray(desired_d).astype(dtype)
+    _assert_allclose_sparse(L, desired_L, atol=1e-12)
+    _assert_allclose_sparse(d, desired_d, atol=1e-12)
+
+    if not copy:
+        if not (normed and check_int_type(mat)):
+            if type(mat) is np.ndarray:
+                assert_allclose(L, mat)
+            elif mat.format == 'coo':
+                _assert_allclose_sparse(L, mat)
+
+
+INT_DTYPES = (np.intc, np_long, np.longlong)
+REAL_DTYPES = (np.float32, np.float64, np.longdouble)
+COMPLEX_DTYPES = (np.complex64, np.complex128, np.clongdouble)
+DTYPES = INT_DTYPES + REAL_DTYPES + COMPLEX_DTYPES
+
+
+@pytest.mark.parametrize("dtype", DTYPES)
+@pytest.mark.parametrize("arr_type", [np.array,
+                                      sparse.csr_matrix,
+                                      sparse.coo_matrix,
+                                      sparse.csr_array,
+                                      sparse.coo_array])
+@pytest.mark.parametrize("copy", [True, False])
+@pytest.mark.parametrize("normed", [True, False])
+@pytest.mark.parametrize("use_out_degree", [True, False])
+def test_asymmetric_laplacian(use_out_degree, normed,
+                              copy, dtype, arr_type):
+    # adjacency matrix
+    A = [[0, 1, 0],
+         [4, 2, 0],
+         [0, 0, 0]]
+    A = arr_type(np.array(A), dtype=dtype)
+    A_copy = A.copy()
+
+    if not normed and use_out_degree:
+        # Laplacian matrix using out-degree
+        L = [[1, -1, 0],
+             [-4, 4, 0],
+             [0, 0, 0]]
+        d = [1, 4, 0]
+
+    if normed and use_out_degree:
+        # normalized Laplacian matrix using out-degree
+        L = [[1, -0.5, 0],
+             [-2, 1, 0],
+             [0, 0, 0]]
+        d = [1, 2, 1]
+
+    if not normed and not use_out_degree:
+        # Laplacian matrix using in-degree
+        L = [[4, -1, 0],
+             [-4, 1, 0],
+             [0, 0, 0]]
+        d = [4, 1, 0]
+
+    if normed and not use_out_degree:
+        # normalized Laplacian matrix using in-degree
+        L = [[1, -0.5, 0],
+             [-2, 1, 0],
+             [0, 0, 0]]
+        d = [2, 1, 1]
+
+    _check_laplacian_dtype_none(
+        A,
+        L,
+        d,
+        normed=normed,
+        use_out_degree=use_out_degree,
+        copy=copy,
+        dtype=dtype,
+        arr_type=arr_type,
+    )
+
+    _check_laplacian_dtype(
+        A_copy,
+        L,
+        d,
+        normed=normed,
+        use_out_degree=use_out_degree,
+        copy=copy,
+        dtype=dtype,
+        arr_type=arr_type,
+    )
+
+
+@pytest.mark.parametrize("fmt", ['csr', 'csc', 'coo', 'lil',
+                                 'dok', 'dia', 'bsr'])
+@pytest.mark.parametrize("normed", [True, False])
+@pytest.mark.parametrize("copy", [True, False])
+def test_sparse_formats(fmt, normed, copy):
+    mat = sparse.diags_array([1, 1], offsets=[-1, 1], shape=(4, 4), format=fmt)
+    _check_symmetric_graph_laplacian(mat, normed, copy)
+
+
+@pytest.mark.parametrize(
+    "arr_type", [np.asarray,
+                 sparse.csr_matrix,
+                 sparse.coo_matrix,
+                 sparse.csr_array,
+                 sparse.coo_array]
+)
+@pytest.mark.parametrize("form", ["array", "function", "lo"])
+def test_laplacian_symmetrized(arr_type, form):
+    # adjacency matrix
+    n = 3
+    mat = arr_type(np.arange(n * n).reshape(n, n))
+    L_in, d_in = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        form=form,
+    )
+    L_out, d_out = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        use_out_degree=True,
+        form=form,
+    )
+    Ls, ds = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        symmetrized=True,
+        form=form,
+    )
+    Ls_normed, ds_normed = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        symmetrized=True,
+        normed=True,
+        form=form,
+    )
+    mat += mat.T
+    Lss, dss = csgraph.laplacian(mat, return_diag=True, form=form)
+    Lss_normed, dss_normed = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        normed=True,
+        form=form,
+    )
+
+    assert_allclose(ds, d_in + d_out)
+    assert_allclose(ds, dss)
+    assert_allclose(ds_normed, dss_normed)
+
+    d = {}
+    for L in ["L_in", "L_out", "Ls", "Ls_normed", "Lss", "Lss_normed"]:
+        if form == "array":
+            d[L] = eval(L)
+        else:
+            d[L] = eval(L)(np.eye(n, dtype=mat.dtype))
+
+    _assert_allclose_sparse(d["Ls"], d["L_in"] + d["L_out"].T)
+    _assert_allclose_sparse(d["Ls"], d["Lss"])
+    _assert_allclose_sparse(d["Ls_normed"], d["Lss_normed"])
+
+
+@pytest.mark.parametrize(
+    "arr_type", [np.asarray,
+                 sparse.csr_matrix,
+                 sparse.coo_matrix,
+                 sparse.csr_array,
+                 sparse.coo_array]
+)
+@pytest.mark.parametrize("dtype", DTYPES)
+@pytest.mark.parametrize("normed", [True, False])
+@pytest.mark.parametrize("symmetrized", [True, False])
+@pytest.mark.parametrize("use_out_degree", [True, False])
+@pytest.mark.parametrize("form", ["function", "lo"])
+def test_format(dtype, arr_type, normed, symmetrized, use_out_degree, form):
+    n = 3
+    mat = [[0, 1, 0], [4, 2, 0], [0, 0, 0]]
+    mat = arr_type(np.array(mat), dtype=dtype)
+    Lo, do = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        normed=normed,
+        symmetrized=symmetrized,
+        use_out_degree=use_out_degree,
+        dtype=dtype,
+    )
+    La, da = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        normed=normed,
+        symmetrized=symmetrized,
+        use_out_degree=use_out_degree,
+        dtype=dtype,
+        form="array",
+    )
+    assert_allclose(do, da)
+    _assert_allclose_sparse(Lo, La)
+
+    L, d = csgraph.laplacian(
+        mat,
+        return_diag=True,
+        normed=normed,
+        symmetrized=symmetrized,
+        use_out_degree=use_out_degree,
+        dtype=dtype,
+        form=form,
+    )
+    assert_allclose(d, do)
+    assert d.dtype == dtype
+    Lm = L(np.eye(n, dtype=mat.dtype)).astype(dtype)
+    _assert_allclose_sparse(Lm, Lo, rtol=2e-7, atol=2e-7)
+    x = np.arange(6).reshape(3, 2)
+    if not (normed and dtype in INT_DTYPES):
+        assert_allclose(L(x), Lo @ x)
+    else:
+        # Normalized Lo is casted to integer, but L() is not
+        pass
+
+
+def test_format_error_message():
+    with pytest.raises(ValueError, match="Invalid form: 'toto'"):
+        _ = csgraph.laplacian(np.eye(1), form='toto')

scipy/sparse/csgraph/tests/test_matching.py

@@ -0,0 +1,307 @@
+from itertools import product
+
+import numpy as np
+from numpy.testing import assert_array_equal, assert_equal
+import pytest
+
+from scipy.sparse import csr_array, coo_array, diags_array
+from scipy.sparse.csgraph import (
+    maximum_bipartite_matching, min_weight_full_bipartite_matching
+)
+
+
+def test_maximum_bipartite_matching_raises_on_dense_input():
+    with pytest.raises(TypeError):
+        graph = np.array([[0, 1], [0, 0]])
+        maximum_bipartite_matching(graph)
+
+
+def test_maximum_bipartite_matching_empty_graph():
+    graph = csr_array((0, 0))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    expected_matching = np.array([])
+    assert_array_equal(expected_matching, x)
+    assert_array_equal(expected_matching, y)
+
+
+def test_maximum_bipartite_matching_empty_left_partition():
+    graph = csr_array((2, 0))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    assert_array_equal(np.array([]), x)
+    assert_array_equal(np.array([-1, -1]), y)
+
+
+def test_maximum_bipartite_matching_empty_right_partition():
+    graph = csr_array((0, 3))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    assert_array_equal(np.array([-1, -1, -1]), x)
+    assert_array_equal(np.array([]), y)
+
+
+def test_maximum_bipartite_matching_graph_with_no_edges():
+    graph = csr_array((2, 2))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    assert_array_equal(np.array([-1, -1]), x)
+    assert_array_equal(np.array([-1, -1]), y)
+
+
+def test_maximum_bipartite_matching_graph_that_causes_augmentation():
+    # In this graph, column 1 is initially assigned to row 1, but it should be
+    # reassigned to make room for row 2.
+    graph = csr_array([[1, 1], [1, 0]])
+    x = maximum_bipartite_matching(graph, perm_type='column')
+    y = maximum_bipartite_matching(graph, perm_type='row')
+    expected_matching = np.array([1, 0])
+    assert_array_equal(expected_matching, x)
+    assert_array_equal(expected_matching, y)
+
+
+def test_maximum_bipartite_matching_graph_with_more_rows_than_columns():
+    graph = csr_array([[1, 1], [1, 0], [0, 1]])
+    x = maximum_bipartite_matching(graph, perm_type='column')
+    y = maximum_bipartite_matching(graph, perm_type='row')
+    assert_array_equal(np.array([0, -1, 1]), x)
+    assert_array_equal(np.array([0, 2]), y)
+
+
+def test_maximum_bipartite_matching_graph_with_more_columns_than_rows():
+    graph = csr_array([[1, 1, 0], [0, 0, 1]])
+    x = maximum_bipartite_matching(graph, perm_type='column')
+    y = maximum_bipartite_matching(graph, perm_type='row')
+    assert_array_equal(np.array([0, 2]), x)
+    assert_array_equal(np.array([0, -1, 1]), y)
+
+
+def test_maximum_bipartite_matching_explicit_zeros_count_as_edges():
+    data = [0, 0]
+    indices = [1, 0]
+    indptr = [0, 1, 2]
+    graph = csr_array((data, indices, indptr), shape=(2, 2))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    expected_matching = np.array([1, 0])
+    assert_array_equal(expected_matching, x)
+    assert_array_equal(expected_matching, y)
+
+
+def test_maximum_bipartite_matching_feasibility_of_result():
+    # This is a regression test for GitHub issue #11458
+    data = np.ones(50, dtype=int)
+    indices = [11, 12, 19, 22, 23, 5, 22, 3, 8, 10, 5, 6, 11, 12, 13, 5, 13,
+               14, 20, 22, 3, 15, 3, 13, 14, 11, 12, 19, 22, 23, 5, 22, 3, 8,
+               10, 5, 6, 11, 12, 13, 5, 13, 14, 20, 22, 3, 15, 3, 13, 14]
+    indptr = [0, 5, 7, 10, 10, 15, 20, 22, 22, 23, 25, 30, 32, 35, 35, 40, 45,
+              47, 47, 48, 50]
+    graph = csr_array((data, indices, indptr), shape=(20, 25))
+    x = maximum_bipartite_matching(graph, perm_type='row')
+    y = maximum_bipartite_matching(graph, perm_type='column')
+    assert (x != -1).sum() == 13
+    assert (y != -1).sum() == 13
+    # Ensure that each element of the matching is in fact an edge in the graph.
+    for u, v in zip(range(graph.shape[0]), y):
+        if v != -1:
+            assert graph[u, v]
+    for u, v in zip(x, range(graph.shape[1])):
+        if u != -1:
+            assert graph[u, v]
+
+
+def test_matching_large_random_graph_with_one_edge_incident_to_each_vertex():
+    np.random.seed(42)
+    A = diags_array(np.ones(25), offsets=0, format='csr')
+    rand_perm = np.random.permutation(25)
+    rand_perm2 = np.random.permutation(25)
+
+    Rrow = np.arange(25)
+    Rcol = rand_perm
+    Rdata = np.ones(25, dtype=int)
+    Rmat = csr_array((Rdata, (Rrow, Rcol)))
+
+    Crow = rand_perm2
+    Ccol = np.arange(25)
+    Cdata = np.ones(25, dtype=int)
+    Cmat = csr_array((Cdata, (Crow, Ccol)))
+    # Randomly permute identity matrix
+    B = Rmat @ A @ Cmat
+
+    # Row permute
+    perm = maximum_bipartite_matching(B, perm_type='row')
+    Rrow = np.arange(25)
+    Rcol = perm
+    Rdata = np.ones(25, dtype=int)
+    Rmat = csr_array((Rdata, (Rrow, Rcol)))
+    C1 = Rmat @ B
+
+    # Column permute
+    perm2 = maximum_bipartite_matching(B, perm_type='column')
+    Crow = perm2
+    Ccol = np.arange(25)
+    Cdata = np.ones(25, dtype=int)
+    Cmat = csr_array((Cdata, (Crow, Ccol)))
+    C2 = B @ Cmat
+
+    # Should get identity matrix back
+    assert_equal(any(C1.diagonal() == 0), False)
+    assert_equal(any(C2.diagonal() == 0), False)
+
+
+@pytest.mark.parametrize('num_rows,num_cols', [(0, 0), (2, 0), (0, 3)])
+def test_min_weight_full_matching_trivial_graph(num_rows, num_cols):
+    biadjacency = csr_array((num_cols, num_rows))
+    biadjacency1 = coo_array((num_cols, num_rows))
+
+    row_ind, col_ind = min_weight_full_bipartite_matching(biadjacency)
+    assert len(row_ind) == 0
+    assert len(col_ind) == 0
+
+    row_ind1, col_ind1 = min_weight_full_bipartite_matching(biadjacency1)
+    assert len(row_ind1) == 0
+    assert len(col_ind1) == 0
+
+
+@pytest.mark.parametrize('biadjacency',
+                         [
+                            [[1, 1, 1], [1, 0, 0], [1, 0, 0]],
+                            [[1, 1, 1], [0, 0, 1], [0, 0, 1]],
+                            [[1, 0, 0, 1], [1, 1, 0, 1], [0, 0, 0, 0]],
+                            [[1, 0, 0], [2, 0, 0]],
+                            [[0, 1, 0], [0, 2, 0]],
+                            [[1, 0], [2, 0], [5, 0]]
+                         ])
+def test_min_weight_full_matching_infeasible_problems(biadjacency):
+    with pytest.raises(ValueError):
+        min_weight_full_bipartite_matching(csr_array(biadjacency))
+    with pytest.raises(ValueError):
+        min_weight_full_bipartite_matching(coo_array(biadjacency))
+
+
+def test_min_weight_full_matching_large_infeasible():
+    # Regression test for GitHub issue #17269
+    a = np.asarray([
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.001],
+        [0.0, 0.11687445, 0.0, 0.0, 0.01319788, 0.07509257, 0.0,
+         0.0, 0.0, 0.74228317, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.81087935, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.8408466, 0.0, 0.0, 0.0, 0.0, 0.01194389,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.82994211, 0.0, 0.0, 0.0, 0.11468516, 0.0, 0.0, 0.0,
+         0.11173505, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.0, 0.0],
+        [0.18796507, 0.0, 0.04002318, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75883335,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.71545464, 0.0, 0.0, 0.0, 0.0, 0.0, 0.02748488,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.78470564, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.14829198,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.10870609, 0.0, 0.0, 0.0, 0.8918677, 0.0, 0.0, 0.0, 0.06306644,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
+         0.63844085, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.7442354, 0.0, 0.0, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.09850549, 0.0, 0.0, 0.18638258,
+         0.2769244, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.73182464, 0.0, 0.0, 0.46443561,
+         0.38589284, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+        [0.29510278, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.09666032, 0.0,
+         0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
+        ])
+    with pytest.raises(ValueError, match='no full matching exists'):
+        min_weight_full_bipartite_matching(csr_array(a))
+    with pytest.raises(ValueError, match='no full matching exists'):
+        min_weight_full_bipartite_matching(coo_array(a))
+
+
+@pytest.mark.thread_unsafe
+def test_explicit_zero_causes_warning():
+    biadjacency = csr_array(((2, 0, 3), (0, 1, 1), (0, 2, 3)))
+    with pytest.warns(UserWarning):
+        min_weight_full_bipartite_matching(biadjacency)
+    with pytest.warns(UserWarning):
+        min_weight_full_bipartite_matching(biadjacency.tocoo())
+
+
+# General test for linear sum assignment solvers to make it possible to rely
+# on the same tests for scipy.optimize.linear_sum_assignment.
+def linear_sum_assignment_assertions(
+    solver, array_type, sign, test_case
+):
+    cost_matrix, expected_cost = test_case
+    maximize = sign == -1
+    cost_matrix = sign * array_type(cost_matrix)
+    expected_cost = sign * np.array(expected_cost)
+
+    row_ind, col_ind = solver(cost_matrix, maximize=maximize)
+    assert_array_equal(row_ind, np.sort(row_ind))
+    assert_array_equal(expected_cost,
+                       np.array(cost_matrix[row_ind, col_ind]).flatten())
+
+    cost_matrix = cost_matrix.T
+    row_ind, col_ind = solver(cost_matrix, maximize=maximize)
+    assert_array_equal(row_ind, np.sort(row_ind))
+    assert_array_equal(np.sort(expected_cost),
+                       np.sort(np.array(
+                           cost_matrix[row_ind, col_ind])).flatten())
+
+
+linear_sum_assignment_test_cases = product(
+    [-1, 1],
+    [
+        # Square
+        ([[400, 150, 400],
+          [400, 450, 600],
+          [300, 225, 300]],
+         [150, 400, 300]),
+
+        # Rectangular variant
+        ([[400, 150, 400, 1],
+          [400, 450, 600, 2],
+          [300, 225, 300, 3]],
+         [150, 2, 300]),
+
+        ([[10, 10, 8],
+          [9, 8, 1],
+          [9, 7, 4]],
+         [10, 1, 7]),
+
+        # Square
+        ([[10, 10, 8, 11],
+          [9, 8, 1, 1],
+          [9, 7, 4, 10]],
+         [10, 1, 4]),
+
+        # Rectangular variant
+        ([[10, float("inf"), float("inf")],
+          [float("inf"), float("inf"), 1],
+          [float("inf"), 7, float("inf")]],
+         [10, 1, 7])
+    ])
+
+
+@pytest.mark.parametrize('sign,test_case', linear_sum_assignment_test_cases)
+def test_min_weight_full_matching_small_inputs(sign, test_case):
+    linear_sum_assignment_assertions(
+        min_weight_full_bipartite_matching, csr_array, sign, test_case)