-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathtest_rescal_fro_mu.py
111 lines (99 loc) · 4.29 KB
/
test_rescal_fro_mu.py
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
import numpy as np
import scipy
import scipy.sparse
from TELF.factorization.decompositions import rescal_fro_mu
from TELF.factorization.decompositions.utilities.math_utils import fro_norm
from TELF.factorization.decompositions.utilities.resample import uniform_product
import pytest
def test_A_update_numpy():
np.random.seed(0)
n, k, t = 4, 2, 3
A0 = np.random.rand(n, k)
R0 = [np.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [np.array, scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
A = rescal_fro_mu.A_update(X, uniform_product(A0, 0.1), R0, use_gpu=False)
assert A.dtype == dtype
# mu update for A regularly fails to converge given a perfect X and R, this is not a issue in practice
#assert np.allclose(A,A0,rtol=1e-2,atol=1e-2)
def test_R_update_numpy():
np.random.seed(0)
n, k, t = 4, 2, 3
A0 = np.random.rand(n, k)
R0 = [np.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [np.array, scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
R = rescal_fro_mu.R_update(
X, A0, [uniform_product(r, 0.1) for r in R0], use_gpu=False)
for r, r0 in zip(R, R0):
assert r.dtype == dtype
assert np.allclose(r, r0, rtol=1e-2, atol=1e-1)
def test_rescal_numpy():
np.random.seed(0)
n, k, t = 4, 2, 3
A0 = np.random.rand(n, k)
R0 = [np.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [np.array, scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
A = uniform_product(A0, 0.1)
R = [uniform_product(r, 0.1) for r in R0]
A, R = rescal_fro_mu.rescal(X, A, R, use_gpu=False)
assert A.dtype == dtype
for x, r in zip(X, R):
assert r.dtype == dtype
assert fro_norm(x-A@[email protected])/fro_norm(x) < 1e-2
def test_A_update_cupy():
cp = pytest.importorskip("cupy")
cupyx = pytest.importorskip("cupyx")
cp.random.seed(0)
n, k, t = 4, 2, 3
A0 = cp.random.rand(n, k)
R0 = [cp.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [cp.array, cupyx.scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
A = rescal_fro_mu.A_update(X, uniform_product(A0, 0.1, use_gpu=True), R0, use_gpu=True)
assert A.dtype == dtype
# mu update for A regularly fails to converge given a perfect X and R, this is not a issue in practice
#assert np.allclose(A,A0,rtol=1e-2,atol=1e-2)
def test_R_update_cupy():
cp = pytest.importorskip("cupy")
cupyx = pytest.importorskip("cupyx")
cp.random.seed(0)
n, k, t = 4, 2, 3
A0 = cp.random.rand(n, k)
R0 = [cp.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [cp.array, cupyx.scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
R = rescal_fro_mu.R_update(
X, A0, [uniform_product(r, 0.1, use_gpu=True) for r in R0], use_gpu=True)
for r, r0 in zip(R, R0):
assert r.dtype == dtype
assert cp.allclose(r, r0, rtol=1e-2, atol=1e-2)
def test_rescal_cupy():
cp = pytest.importorskip("cupy")
cupyx = pytest.importorskip("cupyx")
cp.random.seed(0)
n, k, t = 4, 2, 3
A0 = cp.random.rand(n, k)
R0 = [cp.random.rand(k, k) for _ in range(t)]
X0 = [A0@[email protected] for r in R0]
for dtype in [np.float32, np.float64]:
for typ in [cp.array, cupyx.scipy.sparse.csr_matrix]:
X = [typ(x.astype(dtype)) for x in X0]
A = uniform_product(A0, 0.1, use_gpu=True)
R = [uniform_product(r, 0.1, use_gpu=True) for r in R0]
A, R = rescal_fro_mu.rescal(X, A, R, use_gpu=True)
assert A.dtype == dtype
for x, r in zip(X, R):
assert r.dtype == dtype
assert fro_norm(x-A@[email protected], use_gpu=True)/fro_norm(x, use_gpu=True) < 1e-3