-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
176 lines (152 loc) · 5.43 KB
/
train.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
'''
SeaNMF Training
'''
import os
import time
import argparse
import numpy as np
from utils import *
from model import *
def train(file_suffix="", method="seanmf", alpha=0.1, beta=0, n_topics=10, max_iter=500, max_err=0.1, fix_random=False):
docs = read_docs("data/doc_term_mat_" + file_suffix + ".txt")
vocab = read_vocab("data/vocab_" + file_suffix + ".txt")
n_docs = len(docs)
n_terms = len(vocab)
print('n_docs={}, n_terms={}'.format(n_docs, n_terms))
tmp_folder = 'seanmf_results'
if not os.access(tmp_folder, os.F_OK):
os.mkdir(tmp_folder)
if method.lower() == 'nmf':
print('read term doc matrix')
dt_mat = np.zeros([n_terms, n_docs])
for k in range(n_docs):
for j in docs[k]:
dt_mat[j, k] += 1.0
print('term doc matrix done')
print('-' * 50)
model = NMF(
dt_mat,
n_topics,
max_iter,
max_err)
model.save_format(
Wfile=tmp_folder + '/W_' + file_suffix + '.txt',
Hfile=tmp_folder + '/H_' + file_suffix + '.txt')
if method.lower() == 'seanmf':
print('calculate co-occurence matrix')
dt_mat = np.zeros([n_terms, n_terms])
for itm in docs:
for kk in itm:
for jj in itm:
dt_mat[int(kk), int(jj)] += 1.0
print('co-occur done')
print('-' * 50)
print('calculate PPMI')
D1 = np.sum(dt_mat)
SS = D1 * dt_mat
for k in range(n_terms):
SS[k] /= np.sum(dt_mat[k])
for k in range(n_terms):
SS[:, k] /= np.sum(dt_mat[:, k])
dt_mat = [] # release memory
SS[SS == 0] = 1.0
SS = np.log(SS)
SS[SS < 0.0] = 0.0
print('PPMI done')
print('-' * 50)
print('read term doc matrix')
dt_mat = np.zeros([n_terms, n_docs])
for k in range(n_docs):
for j in docs[k]:
dt_mat[j, k] += 1.0
print('term doc matrix done')
print('-' * 50)
model = SeaNMFL1(
dt_mat, SS,
alpha=alpha,
beta=beta,
n_topic=n_topics,
max_iter=max_iter,
max_err=max_err,
fix_seed=fix_random)
model.save_format(
W1file=tmp_folder + '/W_' + file_suffix + '.txt',
W2file=tmp_folder + '/Wc_' + file_suffix + '.txt',
Hfile=tmp_folder + '/H_' + file_suffix + '.txt')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--corpus_file', default='data/doc_term_mat.txt', help='term document matrix file')
parser.add_argument('--vocab_file', default='data/vocab.txt', help='vocab file')
parser.add_argument('--model', default='seanmf', help='nmf | seanmf')
parser.add_argument('--max_iter', type=int, default=500, help='max number of iterations')
parser.add_argument('--n_topics', type=int, default=100, help='number of topics')
parser.add_argument('--alpha', type=float, default=0.1, help='alpha')
parser.add_argument('--beta', type=float, default=0.0, help='beta')
parser.add_argument('--max_err', type=float, default=0.1, help='stop criterion')
parser.add_argument('--fix_seed', type=bool, default=True, help='set random seed 0')
args = parser.parse_args()
docs = read_docs(args.corpus_file)
vocab = read_vocab(args.vocab_file)
n_docs = len(docs)
n_terms = len(vocab)
print('n_docs={}, n_terms={}'.format(n_docs, n_terms))
tmp_folder = 'seanmf_results'
if not os.access(tmp_folder, os.F_OK):
os.mkdir(tmp_folder)
if args.model.lower() == 'nmf':
print('read term doc matrix')
dt_mat = np.zeros([n_terms, n_docs])
for k in range(n_docs):
for j in docs[k]:
dt_mat[j, k] += 1.0
print('term doc matrix done')
print('-'*50)
model = NMF(
dt_mat,
n_topic=args.n_topics,
max_iter=args.max_iter,
max_err=args.max_err)
model.save_format(
Wfile=tmp_folder+'/W.txt',
Hfile=tmp_folder+'/H.txt')
if args.model.lower() == 'seanmf':
print('calculate co-occurence matrix')
dt_mat = np.zeros([n_terms, n_terms])
for itm in docs:
for kk in itm:
for jj in itm:
dt_mat[int(kk),int(jj)] += 1.0
print('co-occur done')
print('-'*50)
print('calculate PPMI')
D1 = np.sum(dt_mat)
SS = D1*dt_mat
for k in range(n_terms):
SS[k] /= np.sum(dt_mat[k])
for k in range(n_terms):
SS[:,k] /= np.sum(dt_mat[:,k])
dt_mat = [] # release memory
SS[SS==0] = 1.0
SS = np.log(SS)
SS[SS<0.0] = 0.0
print('PPMI done')
print('-'*50)
print('read term doc matrix')
dt_mat = np.zeros([n_terms, n_docs])
for k in range(n_docs):
for j in docs[k]:
dt_mat[j, k] += 1.0
print('term doc matrix done')
print('-'*50)
model = SeaNMFL1(
dt_mat, SS,
alpha=args.alpha,
beta=args.beta,
n_topic=args.n_topics,
max_iter=args.max_iter,
max_err=args.max_err,
fix_seed=args.fix_seed)
model.save_format(
W1file=tmp_folder+'/W.txt',
W2file=tmp_folder+'/Wc.txt',
Hfile=tmp_folder+'/H.txt')