-
Notifications
You must be signed in to change notification settings - Fork 30
/
Copy pathtest_transforms.py
137 lines (113 loc) · 4.2 KB
/
test_transforms.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
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
import os
import numpy as np
import torch
from grelu.transforms.label_transforms import LabelTransform
from grelu.transforms.prediction_transforms import Aggregate, Specificity
from grelu.transforms.seq_transforms import MotifScore, PatternScore
label = np.expand_dims(
np.array([[-1, 101, 30], [3, 0.3, 1000]], dtype=np.float32), 2
) # 2, 3, 1
seqs = ["CAATCGGGAA", "AACGCGCTT", "CTCGTTTCTA"]
cwd = os.path.realpath(os.path.dirname(__file__))
meme_file = os.path.join(cwd, "files", "test.meme")
preds = torch.rand(2, 4, 6)
def test_label_transform():
# Threshold + log
t = LabelTransform(min_clip=0, max_clip=100, transform_func=np.log)
assert np.allclose(t(label).squeeze(), np.log([[0, 100, 30], [3, 0.3, 100]]))
# Threshold only
t = LabelTransform(min_clip=0, max_clip=100, transform_func=None)
assert np.allclose(t(label).squeeze(), np.array([[0, 100, 30], [3, 0.3, 100]]))
def test_pattern_score():
t = PatternScore(patterns=["AA", "TT"], weights=[1, -0.5])
assert np.allclose(t(seqs), np.array([2, 0.5, -1]))
def test_motif_score():
t = MotifScore(motifs=meme_file, weights=[-1, 0.5], rc=False)
assert t(seqs) == [0, 0, 0]
assert t(["CCCACGTGAA", "AATGCGTGGG"]) == [-1, 0.5]
def test_aggregate():
t = Aggregate(
tasks=[0, 1], positions=[3, 4], length_aggfunc="sum", task_aggfunc="sum"
)
expected = preds[:, [0, 1], :][:, :, [3, 4]].sum(axis=(1, 2), keepdims=True)
assert torch.allclose(t(preds), expected)
assert np.allclose(t.compute(preds.numpy()), expected.numpy())
t = Aggregate(tasks=[0, 1], length_aggfunc=None, task_aggfunc="mean")
expected = preds[:, [0, 1], :].mean(1, keepdims=True)
assert torch.allclose(t(preds), expected)
assert np.allclose(t.compute(preds.numpy()), expected.numpy())
t = Aggregate(task_aggfunc=None, length_aggfunc="sum")
expected = preds.sum(2, keepdims=True)
assert torch.allclose(t(preds), expected)
assert np.allclose(t.compute(preds.numpy()), expected.numpy())
t = Aggregate(task_aggfunc=None, length_aggfunc="sum", weight=-1)
expected = -1 * preds.sum(2, keepdims=True)
assert np.allclose(t.compute(preds.numpy()), expected.numpy())
assert torch.allclose(t(preds), expected)
def test_specificity():
# No weight
t = Specificity(
on_tasks=[0],
off_tasks=[1, 2],
on_aggfunc="mean",
off_aggfunc="max",
compare_func="divide",
)
expected_on = preds[:, [0], :].numpy().sum(2, keepdims=True)
expected_off = (
preds[:, [1, 2], :].numpy().sum(2, keepdims=True).max(1, keepdims=True)
)
out = t(preds).numpy()
assert np.allclose(out, np.divide(expected_on, expected_off))
# Constant weight
t = Specificity(
on_tasks=[0],
off_tasks=[1, 2],
on_aggfunc="mean",
off_aggfunc="max",
compare_func="divide",
off_weight=2,
)
expected_on = preds[:, [0], :].numpy().sum(2, keepdims=True)
expected_off = (
preds[:, [1, 2], :].numpy().sum(2, keepdims=True).max(1, keepdims=True)
)
expected_off = expected_off * 2
out = t(preds).numpy()
assert np.allclose(out, np.divide(expected_on, expected_off))
def test_specificity_threshold():
# Thresholded weight
t = Specificity(
on_tasks=[0],
off_tasks=[1],
on_aggfunc="mean",
off_aggfunc="max",
compare_func="divide",
off_weight=2,
off_thresh=1,
positions=[1, 2],
)
preds = torch.Tensor(
[
[0.1, 0.2, 1, 1.2],
[0, -1, 10, 10],
]
).unsqueeze(
0
) # 1, 2, 4
expected_on = np.expand_dims(np.expand_dims(np.array([1.2]), 0), 2)
expected_off = np.expand_dims(np.expand_dims(np.array([18]), 0), 2)
out = t(preds).numpy()
assert np.allclose(out, np.divide(expected_on, expected_off))
preds = torch.Tensor(
[
[0.1, 0.2, 1, 1.2],
[0, -1, 1.1, 10],
]
).unsqueeze(
0
) # 1, 2, 4
expected_on = np.expand_dims(np.expand_dims(np.array([1.2]), 0), 2)
expected_off = np.expand_dims(np.expand_dims(np.array([0.1]), 0), 2)
out = t(preds).numpy()
assert np.allclose(out, np.divide(expected_on, expected_off))