-
Notifications
You must be signed in to change notification settings - Fork 39
/
embedding_eval.py
131 lines (105 loc) · 4.61 KB
/
embedding_eval.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
import numpy as np
from text import VocabBuild
from util import cosines
class EmbeddingEval:
def __init__(self, ground, predictions, embeddings_file, downcase):
self.eval = {}
self.ground = ground
self.predictions = predictions
self.embeddings_file = embeddings_file
self.v = VocabBuild(embeddings_file, downcase=downcase)
self.v.read()
def evaluate(self):
assert len(self.ground) == len(self.predictions)
scorers = [(Average(), "emb-average"),
(Greedy(), "emb-greedy"),
(Extrema(), "emb-extrema")]
for scorer, method in scorers:
score, scores = self.compute_score(scorer, self.ground, self.predictions)
self.eval[method] = score
def compute_score(self, scorer, ground, predictions):
scores = []
for id, ground_ans in ground.items():
predicted_ans = predictions[id]
scores.append(scorer.calculate_score(self.v, ground_ans, predicted_ans))
average_score = np.mean(np.array(scores))
return average_score, np.array(scores)
class Average:
@staticmethod
def calculate_score(v, ground_answers, predicted_ans):
assert type(predicted_ans) is list
assert len(predicted_ans) == 1
assert type(ground_answers) is list
assert len(ground_answers) > 0
ground_answers_means = []
for ground_ans in ground_answers:
ground_ans_idx = v.line_to_seq(ground_ans, output_nan=True)
if not ground_ans_idx:
continue
ground_answers_means.append(v.W[np.array(ground_ans_idx)].mean(axis=0))
assert ground_answers_means
predicted_ans_idx = v.line_to_seq(predicted_ans[0], output_nan=True)
if not predicted_ans_idx:
score = 0
else:
predicted_ans_mean = v.W[np.array(predicted_ans_idx)].mean(axis=0)
score = cosines(np.array(ground_answers_means), predicted_ans_mean).max()
return score
class Extrema:
@staticmethod
def calculate_score(v, ground_answers, predicted_ans):
assert type(predicted_ans) is list
assert len(predicted_ans) == 1
assert type(ground_answers) is list
assert len(ground_answers) > 0
ground_answers_extrema_vecs = []
for ground_ans in ground_answers:
ground_ans_idx = v.line_to_seq(ground_ans, output_nan=True)
if not ground_ans_idx:
continue
ground_ans_vecs = v.W[np.array(ground_ans_idx)]
extrema_arr = abs(ground_ans_vecs).argmax(axis=0)
extrema_vec = ground_ans_vecs[extrema_arr, list(range(len(extrema_arr)))]
ground_answers_extrema_vecs.append(extrema_vec)
assert ground_answers_extrema_vecs
predicted_ans_idx = v.line_to_seq(predicted_ans[0], output_nan=True)
if not predicted_ans_idx:
score = 0
else:
predicted_ans_vecs = v.W[np.array(predicted_ans_idx)]
extrema_arr = abs(predicted_ans_vecs).argmax(axis=0)
predicted_extrema_vec = predicted_ans_vecs[extrema_arr, list(range(len(extrema_arr)))]
score = cosines(np.array(ground_answers_extrema_vecs), predicted_extrema_vec).max()
return score
class Greedy:
def calculate_score(self, v, ground_answers, predicted_ans):
assert type(predicted_ans) is list
assert len(predicted_ans) == 1
assert type(ground_answers) is list
assert len(ground_answers) > 0
scores = []
predicted_ans_idx = v.line_to_seq(predicted_ans[0], output_nan=True)
for ground_ans in ground_answers:
ground_ans_idx = v.line_to_seq(ground_ans, output_nan=True)
if not ground_ans_idx:
continue
# direction 1
if not predicted_ans_idx:
avg_max_cosines_dir1 = 0
else:
max_cosines = []
for id in ground_ans_idx:
max_cosines.append(self.get_max_cosine(v, id, predicted_ans_idx))
avg_max_cosines_dir1 = np.mean(max_cosines)
# direction 2
max_cosines = []
for id in predicted_ans_idx:
max_cosines.append(self.get_max_cosine(v, id, ground_ans_idx))
avg_max_cosines_dir2 = np.mean(max_cosines)
scores.append(np.mean(np.array([avg_max_cosines_dir1, avg_max_cosines_dir2])))
assert scores
return np.max(scores)
@staticmethod
def get_max_cosine(v, id1, ids2):
max_cosine = cosines(v.W[np.array(ids2)], v.W[id1]).max()
return max_cosine