sp_matmul_rs 0.2.1

Sparse CSR matrix multiplication with top-n selection — Rust port of sparse_dot_topn.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261

import sys
from itertools import product

import numpy as np
import pytest
from scipy import sparse

from sp_matmul_rs import sp_matmul, sp_matmul_topn, zip_sp_matmul_topn
from sp_matmul_rs import _rust_parallel_enabled as _has_openmp_support

from ._resources import _assert_array_equal, _assert_smat_equal, _get_topn_elements


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_default(rng, dtype):
    A = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    C = sp_matmul(A, B)
    C_ref = A.dot(B)
    _assert_smat_equal(C, C_ref)


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_nthreads(rng, dtype):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=dtype, random_state=rng)
    C_ref = A.dot(B)
    if not _has_openmp_support:
        with pytest.warns(UserWarning):
            C = sp_matmul(A, B, n_threads=2)
    else:
        C = sp_matmul(A, B, n_threads=2)
    _assert_smat_equal(C, C_ref)


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_default(rng, dtype):
    A = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    C = sp_matmul_topn(A, B, top_n=B.shape[1])
    C_ref = A.dot(B)
    _assert_smat_equal(C, C_ref)


def _sorted_topn(x: np.ndarray, n: int) -> np.ndarray:
    """Top-n values from x, sorted ascending (order-independent comparison)."""
    return np.sort(_get_topn_elements(x, n))


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_topn(rng, dtype):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=dtype, random_state=rng)
    C_ref = A.dot(B)
    C_10 = sp_matmul_topn(A, B, top_n=10)
    C_30 = sp_matmul_topn(A, B, top_n=30)
    for i in range(A.shape[0]):
        _assert_array_equal(np.sort(C_10[i, :].data), _sorted_topn(C_ref[i, :].data, 10))
        _assert_array_equal(np.sort(C_30[i, :].data), _sorted_topn(C_ref[i, :].data, 30))


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_nthreads(rng, dtype):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=dtype, random_state=rng)
    C_ref = A.dot(B)
    if not _has_openmp_support:
        with pytest.warns(UserWarning):
            C = sp_matmul_topn(A, B, top_n=B.shape[1], n_threads=2)
    else:
        C = sp_matmul_topn(A, B, top_n=B.shape[1], n_threads=2)
    _assert_smat_equal(C, C_ref)

    C_10 = sp_matmul_topn(A, B, top_n=10, n_threads=2)
    C_30 = sp_matmul_topn(A, B, top_n=30, n_threads=2)
    for i in range(A.shape[0]):
        _assert_array_equal(np.sort(C_10[i, :].data), _sorted_topn(C_ref[i, :].data, 10))
        _assert_array_equal(np.sort(C_30[i, :].data), _sorted_topn(C_ref[i, :].data, 30))


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_chunk_cols(rng, dtype):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=dtype, random_state=rng)
    C_ref = sp_matmul_topn(A, B, top_n=10)
    for chunk_cols in (1, 7, 64, B.shape[1], 10 * B.shape[1]):
        C = sp_matmul_topn(A, B, top_n=10, chunk_cols=chunk_cols)
        _assert_smat_equal(C, C_ref)
    with pytest.raises(ValueError, match="chunk_cols"):
        sp_matmul_topn(A, B, top_n=10, chunk_cols=0)


def test_sp_matmul_topn_chunk_cols_env(rng, monkeypatch):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=np.float64, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=np.float64, random_state=rng)
    C_ref = sp_matmul_topn(A, B, top_n=10)
    monkeypatch.setenv("SP_MATMUL_RS_CHUNK_COLS", "64")
    C = sp_matmul_topn(A, B, top_n=10)
    _assert_smat_equal(C, C_ref)


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_sorted(rng, dtype):
    A = sparse.random(100, 10, density=0.5, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(10, 100, density=0.5, format="csr", dtype=dtype, random_state=rng)
    C_ref = A.dot(B)

    C_10 = sp_matmul_topn(A, B, top_n=10, sort=True)
    C_30 = sp_matmul_topn(A, B, top_n=30, sort=True, n_threads=2)
    for i in range(A.shape[0]):
        sorted_row = np.sort(C_ref[i, :].data)[::-1]
        _assert_array_equal(C_10[i, :].data, sorted_row[:10])
        _assert_array_equal(C_30[i, :].data, sorted_row[:30])


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_density(rng, dtype):
    A = sparse.random(200, 200, density=0.9, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(200, 200, density=0.9, format="csr", dtype=dtype, random_state=rng)
    C = sp_matmul_topn(A, B, top_n=B.shape[1], density=0.01)
    C_ref = A.dot(B)
    _assert_smat_equal(C, C_ref)


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_threshold(rng, dtype):
    A = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    threshold = 0.0 if np.issubdtype(A.data.dtype, np.integer) else 0.5
    C = sp_matmul_topn(A, B, top_n=B.shape[1], threshold=threshold)
    C_ref = A.dot(B)
    assert C.data.min() > threshold
    assert C.data.size < C_ref.data.size


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_sp_matmul_topn_idx_dtype(rng, dtype):
    A = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(100, 100, density=0.1, format="csr", dtype=dtype, random_state=rng)
    C = sp_matmul_topn(A, B, top_n=B.shape[1], idx_dtype=np.int64)
    C_ref = A.dot(B)
    _assert_smat_equal(C, C_ref)


_FORMATS = ["coo", "csr", "csc"]


@pytest.mark.parametrize("formats", product(_FORMATS, _FORMATS))
def test_sp_matmul_topn_formats(rng, formats):
    A = sparse.random(200, 100, density=0.1, format=formats[0], random_state=rng)
    B = sparse.random(100, 300, density=0.1, format=formats[1], random_state=rng)

    C = sp_matmul_topn(A, B, top_n=B.shape[1])
    C_ref = A.tocsr().dot(B.tocsr())
    _assert_smat_equal(C, C_ref)


def test_sp_matmul_topn_transpose(rng):
    A = sparse.random(200, 100, density=0.1, format="csr", random_state=rng)
    B = sparse.random(100, 300, density=0.1, format="csr", random_state=rng)

    C = sp_matmul_topn(A.transpose(), B.transpose(), top_n=B.shape[1])
    C_ref = A.dot(B)
    _assert_smat_equal(C, C_ref)


_KWARGS = [
    {
        "A": {"m": 10, "n": 10, "density": 0.1, "format": "csr"},
        "B": {"m": 10, "n": 10, "density": 0.1, "format": "csr", "T": False},
    },
    {
        "A": {"m": 100, "n": 100, "density": 0.1, "format": "csr"},
        "B": {"m": 100, "n": 100, "density": 0.1, "format": "csr", "T": False},
    },
    {
        "A": {"m": 1000, "n": 1000, "density": 0.05, "format": "csr"},
        "B": {"m": 1000, "n": 1000, "density": 0.05, "format": "csr", "T": False},
    },
    {
        "A": {"m": 2000, "n": 2000, "density": 0.01, "format": "csr"},
        "B": {"m": 2000, "n": 2000, "density": 0.01, "format": "csr", "T": False},
    },
    {
        "A": {"m": 10000, "n": 10000, "density": 0.001, "format": "csr"},
        "B": {"m": 10000, "n": 10000, "density": 0.001, "format": "csr", "T": False},
    },
    {
        "A": {"m": 1000, "n": 100, "density": 0.1, "format": "csr"},
        "B": {"m": 100, "n": 100, "density": 0.1, "format": "csr", "T": False},
    },
    {
        "A": {"m": 1000, "n": 100, "density": 0.1, "format": "csr"},
        "B": {"m": 100, "n": 20, "density": 0.1, "format": "csr", "T": False},
    },
    {
        "A": {"m": 1000, "n": 100, "density": 0.1, "format": "csr"},
        "B": {"m": 200, "n": 100, "density": 0.1, "format": "csr", "T": True},
    },
    {
        "A": {"m": 1000, "n": 100, "density": 0.1, "format": "csr"},
        "B": {"m": 200, "n": 100, "density": 0.1, "format": "csc", "T": True},
    },
]


@pytest.mark.parametrize("kwargs", _KWARGS)
def test_sp_matmul_topn_broadcasting(rng, kwargs):
    A = sparse.random(**kwargs["A"], random_state=rng)
    T_b = kwargs["B"].pop("T")
    B = sparse.random(**kwargs["B"], random_state=rng)

    topn = B.shape[0] if T_b else B.shape[1]
    C_ref = A.dot(B.T) if T_b else A.dot(B)
    C = sp_matmul_topn(A, B, top_n=topn)
    _assert_smat_equal(C, C_ref)


@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_zip_sp_matmul_topn(rng, dtype):
    # matching 100 names against 600 gt-names, where gt has been split into three parts
    A = sparse.random(100, 2000, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(600, 2000, density=0.1, format="csr", dtype=dtype, random_state=rng)

    # reference
    C_ref = sp_matmul_topn(A, B.T, top_n=10, threshold=0.01, sort=True)

    # zipped C-matrix
    Bs = [B[:100], B[100:300], B[300:]]
    Cs = [sp_matmul_topn(A, Bi.T, top_n=10, threshold=0.01, sort=True) for Bi in Bs]
    C_zip = zip_sp_matmul_topn(top_n=10, C_mats=Cs)

    _assert_array_equal(C_zip.indptr, C_ref.indptr)
    _assert_array_equal(C_zip.data, C_ref.data)
    _assert_array_equal(C_zip.indices, C_ref.indices)


@pytest.mark.skipif(sys.version_info < (3, 9), reason="not all dtypes supported in scipy vstack for python 3.8")
@pytest.mark.parametrize("dtype", [np.float32, np.float64, np.int32, np.int64])
def test_stack_zip_sp_matmul_topn(rng, dtype):
    # matching 1000 names against 600 gt-names,
    # where gt has been split into three parts, and names-to-match in five parts
    A = sparse.random(1000, 2000, density=0.1, format="csr", dtype=dtype, random_state=rng)
    B = sparse.random(600, 2000, density=0.1, format="csr", dtype=dtype, random_state=rng)

    # reference
    C_ref = sp_matmul_topn(A, B.T, top_n=10, threshold=0.01, sort=True)

    # zipped C-matrix
    As = [A[i * 200 : (i + 1) * 200] for i in range(5)]  # names-to-match split
    Bs = [B[:100], B[100:300], B[300:]]  # GT split

    Cs = [[sp_matmul_topn(Aj, Bi.T, top_n=10, threshold=0.01, sort=True) for Bi in Bs] for Aj in As]
    C_zips = [zip_sp_matmul_topn(top_n=10, C_mats=Cis) for Cis in Cs]

    # stack over names-to-match subparts
    C_stack = sparse.vstack(C_zips, dtype=dtype)

    _assert_array_equal(C_stack.indptr, C_ref.indptr)
    _assert_array_equal(C_stack.data, C_ref.data)
    _assert_array_equal(C_stack.indices, C_ref.indices)