-
Notifications
You must be signed in to change notification settings - Fork 0
/
analogy_decomposition.py
181 lines (142 loc) · 6.84 KB
/
analogy_decomposition.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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
import numpy as np
import pandas as pd
import time
import sys
from read_bats import bats_names_pairs
from models import vocabulary_model, load_model
from os.path import exists
from os import mkdir
def start_end_words(model, pairs_sets, vocabulary):
start_words = np.array(
[[model.wv.get_vector(i[0]) for i in pairs_sets[k] if i[0] in vocabulary and i[1] in vocabulary]
for k in range(len(pairs_sets))])
end_words = np.array(
[[model.wv.get_vector(i[1]) for i in pairs_sets[k] if i[0] in vocabulary and i[1] in vocabulary]
for k in range(len(pairs_sets))])
return(start_words, end_words)
def analogy_decomposition(start_words, end_words):
b_bp = []
oa_ob = []
oa_b = []
for i in range(len(start_words)):
b_bp_categ = []
oa_ob_categ = []
oa_b_categ = []
list_start_words = list(start_words[i])
list_end_words = list(end_words[i])
for j in range(len(list_start_words)):
for k in range(len(list_start_words)):
if j != k:
a, ap, b, bp = list_start_words[j], list_end_words[j], list_start_words[k], list_end_words[k]
o_a, o_b = ap - a, bp - b
analogy = b + o_a
norme_analogie_m1 = 1 / (np.linalg.norm(analogy))
norme_bp_m1 = 1 / (np.linalg.norm(bp))
b_bp_categ.append(b @ bp * norme_analogie_m1 * norme_bp_m1)
oa_ob_categ.append(o_a @ o_b * norme_analogie_m1 * norme_bp_m1)
oa_b_categ.append(o_a @ b * norme_analogie_m1 * norme_bp_m1)
b_bp.append(np.mean(b_bp_categ))
oa_ob.append(np.mean(oa_ob_categ))
oa_b.append(np.mean(oa_b_categ))
return([b_bp, oa_ob, oa_b])
def analogy_decomposition_reference(start_words, end_words):
analogy_b = []
oa_b = []
oa2 = []
for i in range(len(start_words)):
analogy_b_categ = []
oa_b_categ = []
oa2_categ = []
list_start_words = list(start_words[i])
list_end_words = list(end_words[i])
for j in range(len(list_start_words)):
for k in range(len(list_start_words)):
if j != k:
a, ap, b, bp = list_start_words[j], list_end_words[j], list_start_words[k], list_end_words[k]
o_a, o_b = ap - a, bp - b
analogy = b + o_a
norme_analogie_m1 = 1 / (np.linalg.norm(analogy))
analogy_b_categ.append((norme_analogie_m1 ** 2) * analogy @ b)
oa_b_categ.append((norme_analogie_m1 ** 2) * o_a @ b)
oa2_categ.append((norme_analogie_m1 ** 2) * o_a @ o_a)
analogy_b.append(np.mean(analogy_b_categ))
oa_b.append(np.mean(oa_b_categ))
oa2.append(np.mean(oa2_categ))
return ([analogy_b, oa_b, oa2])
def delta_sim(start_words, end_words):
analogy_b = []
oa_ob = []
b_ob = []
for i in range(len(start_words)):
analogy_b_categ = []
oa_ob_categ = []
b_ob_categ = []
list_start_words = list(start_words[i])
list_end_words = list(end_words[i])
for j in range(len(list_start_words)):
for k in range(len(list_start_words)):
if j != k:
a, ap, b, bp = list_start_words[j], list_end_words[j], list_start_words[k], list_end_words[k]
o_a, o_b = ap - a, bp - b
analogy = b + o_a
norme_analogie_m1 = 1 / (np.linalg.norm(analogy))
norme_bp_m1 = 1 / (np.linalg.norm(bp))
norme_b_m1 = 1 / (np.linalg.norm(b))
analogy_b_categ.append(norme_analogie_m1 * (norme_bp_m1 - norme_b_m1) * (b @ analogy))
oa_ob_categ.append(norme_analogie_m1 * norme_bp_m1 * o_a @ o_b)
b_ob_categ.append(norme_analogie_m1 * norme_bp_m1 * b @ o_b)
analogy_b.append(np.mean(analogy_b_categ))
oa_ob.append(np.mean(oa_ob_categ))
b_ob.append(np.mean(b_ob_categ))
return ([analogy_b, oa_ob, b_ob])
def all_decompositions(start_words, end_words):
results_decompo = analogy_decomposition(start_words, end_words)
results_decompo_ref = analogy_decomposition_reference(start_words, end_words)
results_delta_sim = delta_sim(start_words, end_words)
return([results_decompo, results_decompo_ref, results_delta_sim])
def decompo(model, pairs_sets, vocabulary, decomposition='all'):
start_words, end_words = start_end_words(model, pairs_sets, vocabulary)
if decomposition=='all':
return(all_decompositions(start_words, end_words))
if decomposition=='decomposition':
return (analogy_decomposition(start_words, end_words))
if decomposition=='decomposition_ref':
return (analogy_decomposition_reference(start_words, end_words))
if decomposition=='delta_sim':
return (delta_sim(start_words, end_words))
def save_decompo(names, results, decomposition):
if decomposition == 'decomposition': columns = np.array(['Categories', 'b*b\'', 'o_a*o_b', 'o_a*b'])
if decomposition == 'decomposition_ref': columns = np.array(['Categories', 'b*analogy', 'o_a*b', 'o_a^2'])
if decomposition == 'delta_sim': columns = np.array(['Categories', 'rest', 'o_a*o_b', 'o_b*b'])
r = np.array([names,results[0],results[1],results[2]])
df = pd.DataFrame(r.T, columns=columns)
if not exists('results'):
print("# ", str('results'), "not found, creating dir.")
mkdir('results')
timestr = time.strftime("%Y%m%d-%H%M%S")
namepath = 'results/' + str(decomposition) + '-' + str(timestr) + '.csv'
df.to_csv(namepath, index=False)
print("# Successfully saved the decomposition to ", str(namepath))
if __name__ == "__main__":
# execute only if run as a script
if len(sys.argv) < 2:
raise("# Please provide a model (name, or filename for a custom model)")
if len(sys.argv) < 3:
raise("# Please provide a decomposition type (decomposition, decomposition_ref, delta_sim or all)")
model_name = str(sys.argv[1])
decomposition = str(sys.argv[2])
print("# Model: ", model_name, " ; Decomposition: ", decomposition)
names, pairs_sets = bats_names_pairs(dir="BATS_3.0")
print('# Loading model for decompositions')
model = load_model(model_name)
vocabulary = vocabulary_model(model)
print('# Computing the decompositions')
results = decompo(model, pairs_sets, vocabulary, decomposition=decomposition)
print('# Saving the decompositions')
if decomposition == 'all':
save_decompo(names, results[0], 'decomposition')
save_decompo(names, results[1], 'decomposition_ref')
save_decompo(names, results[2], 'delta_sim')
else:
save_decompo(names, results, decomposition)
print("# Successfully computed the wanted decomposition of the model ", model_name)