-
Notifications
You must be signed in to change notification settings - Fork 0
/
codenames.py
144 lines (103 loc) · 3.88 KB
/
codenames.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
138
139
140
141
142
import argparse
import gensim
import gensim.downloader
import pickle
import gensim.models
from gensim.models.keyedvectors import KeyedVectors
import numpy as np
import re
def filter_word(word, pos, neg):
if re.match(r'[^a-z]', word):
return False
if len(word) <= 1:
return False
if word in pos:
return False
if word in neg:
return False
return True
def serialize_models():
model_names = ['fasttext-wiki-news-subwords-300', 'glove-twitter-100',
'glove-wiki-gigaword-100', 'word2vec-google-news-300', 'word2vec-ruscorpora-300']
for mod in model_names:
model = gensim.downloader.load(mod)
valid_words = []
for i in range(len(model.index_to_key)):
if filter_word(model.index_to_key[i], [], []):
valid_words.append(model.index_to_key[i])
with open(f'{mod}.pickle', 'wb') as outfile:
pickle.dump(model, outfile)
outfile.close()
with open(f'{mod}_words.pickle', 'wb') as outfile2:
pickle.dump(valid_words, outfile2)
outfile2.close()
parser = argparse.ArgumentParser()
parser.add_argument('--model', required=False, default = 0)
parser.add_argument('--pos', required=True)
parser.add_argument('--neg', required=False, default="")
parser.add_argument("--n_words", required=False, default=10)
parser.add_argument("--alpha", required=False, default=1)
parser.add_argument("--beta", required=False, default=0.5)
parser.add_argument("--serialize", required=True, default=0.5)
def distance_matrix(words, model):
l2_arr = np.ndarray(shape=(len(words), len(words)))
cos_arr = np.ndarray(shape=(len(words), len(words)))
for i in range(len(words)):
for j in range(len(words)):
w_i = model[words[i]]
w_j = model[words[j]]
l2_dist = np.linalg.norm(w_i - w_j)
l2_arr[i][j] = round(l2_dist)
cos_arr[i][j] = round(model.distance(words[i], words[j]), ndigits=3)
return l2_arr, cos_arr
w1 = ["car", "gasoline", "alcohol", "party"]
w2 = ["tablet", "battery", "house", "california", "solar"]
w3 = ["hat", "fruit", "computer", "pencil", "dog"]
def generate_clue(pos, neg = [], n_words=10, alpha = 1, beta = 0.5, model_num=0):
model_name = model_names[model_num]
with open(f"{model_name}.pickle", "rb") as infile:
model: KeyedVectors = pickle.load(infile)
infile.close()
with open(f'{model_name}_words.pickle', 'rb') as infile2:
valid_words = pickle.load(infile2)
infile2.close()
pos_vectors = list(map(lambda x: model[x], pos))
neg_vectors = list(map(lambda x: model[x], neg))
best = [("", float('inf'))]*n_words
for word in valid_words:
if filter_word(word, pos, neg):
vec = model[word]
loss = compute_loss(vec, pos_vectors, neg_vectors, alpha, beta)
if loss < best[-1][1]:
best[-1] = (word, loss)
best.sort(key=lambda x:x[1])
return best
def compute_loss(vec, pos, neg, alpha, beta):
running_loss = 0
for v in pos:
#dist = np.linalg.norm(v - vec)
dist = 1 - cosine_similarity(v, vec)
running_loss += alpha*(dist**2)
for v in neg:
#dist = np.linalg.norm(v - vec)
dist = 1 - cosine_similarity(v, vec)
running_loss -= beta*(dist**2)
return running_loss
def cosine_similarity(v1, v2):
dot = np.dot(v1, v2)
n1 = np.linalg.norm(v1)
n2 = np.linalg.norm(v2)
return dot/(n1*n2)
if __name__ == '__main__':
args = parser.parse_args()
model_num = int(args.model)
pos = args.pos
neg = args.neg
n_words = int(args.n_words)
alpha = args.alpha
beta = args.beta
p = pos.split()
n = neg.split()
if args.serialize:
serialize_models()
print(generate_clue(pos=p, neg=n, n_words=n_words, alpha=alpha, beta=beta, model_num=model_num))