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
|
"""Tests for tensorflow.kernels.edit_distance_op."""
import tensorflow.python.platform
import numpy as np
import tensorflow as tf
def ConstantOf(x):
x = np.asarray(x)
# Convert to int64 if it's not a string
if x.dtype.char != "S": x = np.asarray(x, dtype=np.int64)
return tf.constant(x)
class EditDistanceTest(tf.test.TestCase):
def _testEditDistance(self, hypothesis, truth, normalize,
expected_output, expected_err_re=None):
# hypothesis and truth are (index, value, shape) tuples
hypothesis_st = tf.SparseTensor(*[ConstantOf(x) for x in hypothesis])
truth_st = tf.SparseTensor(*[ConstantOf(x) for x in truth])
edit_distance = tf.edit_distance(
hypothesis=hypothesis_st, truth=truth_st, normalize=normalize)
with self.test_session():
if expected_err_re is None:
# Shape inference figures out the shape from the shape variables
expected_shape = [
max(h, t) for h, t in zip(hypothesis[2], truth[2])[:-1]]
self.assertEqual(edit_distance.get_shape(), expected_shape)
output = edit_distance.eval()
self.assertAllClose(output, expected_output)
else:
with self.assertRaisesOpError(expected_err_re):
edit_distance.eval()
def testEditDistanceNormalized(self):
hypothesis_indices = [[0, 0], [0, 1],
[1, 0], [1, 1]]
hypothesis_values = [0, 1,
1, -1]
hypothesis_shape = [2, 2]
truth_indices = [[0, 0],
[1, 0], [1, 1]]
truth_values = [0,
1, 1]
truth_shape = [2, 2]
expected_output = [1.0, 0.5]
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=True,
expected_output=expected_output)
def testEditDistanceUnnormalized(self):
hypothesis_indices = [[0, 0],
[1, 0], [1, 1]]
hypothesis_values = [10,
10, 11]
hypothesis_shape = [2, 2]
truth_indices = [[0, 0], [0, 1],
[1, 0], [1, 1]]
truth_values = [1, 2,
1, -1]
truth_shape = [2, 3]
expected_output = [2.0, 2.0]
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=False,
expected_output=expected_output)
def testEditDistanceProperDistance(self):
# In this case, the values are individual characters stored in the
# SparseTensor (type DT_STRING)
hypothesis_indices = ([[0, i] for i, _ in enumerate("algorithm")] +
[[1, i] for i, _ in enumerate("altruistic")])
hypothesis_values = [x for x in "algorithm"] + [x for x in "altruistic"]
hypothesis_shape = [2, 11]
truth_indices = ([[0, i] for i, _ in enumerate("altruistic")] +
[[1, i] for i, _ in enumerate("algorithm")])
truth_values = [x for x in "altruistic"] + [x for x in "algorithm"]
truth_shape = [2, 11]
expected_unnormalized = [6.0, 6.0]
expected_normalized = [6.0/len("altruistic"),
6.0/len("algorithm")]
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=False,
expected_output=expected_unnormalized)
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=True,
expected_output=expected_normalized)
def testEditDistance3D(self):
hypothesis_indices = [[0, 0, 0],
[1, 0, 0]]
hypothesis_values = [0, 1]
hypothesis_shape = [2, 1, 1]
truth_indices = [[0, 1, 0],
[1, 0, 0],
[1, 1, 0]]
truth_values = [0, 1, 1]
truth_shape = [2, 2, 1]
expected_output = [[np.inf, 1.0], # (0,0): no truth, (0,1): no hypothesis
[0.0, 1.0]] # (1,0): match, (1,1): no hypothesis
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=True,
expected_output=expected_output)
def testEditDistanceMissingHypothesis(self):
hypothesis_indices = np.empty((0, 2), dtype=np.int64)
hypothesis_values = []
hypothesis_shape = [1, 0]
truth_indices = [[0, 0]]
truth_values = [0]
truth_shape = [1, 1]
expected_output = [1.0]
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=True,
expected_output=expected_output)
def testEditDistanceMissingTruth(self):
hypothesis_indices = [[0, 0]]
hypothesis_values = [0]
hypothesis_shape = [1, 1]
truth_indices = np.empty((0, 2), dtype=np.int64)
truth_values = []
truth_shape = [1, 0]
expected_output = [np.inf] # Normalized, divide by zero
self._testEditDistance(
hypothesis=(hypothesis_indices, hypothesis_values, hypothesis_shape),
truth=(truth_indices, truth_values, truth_shape),
normalize=True,
expected_output=expected_output)
if __name__ == "__main__":
tf.test.main()
|
