/
skipgram-trainer.py
executable file
·294 lines (240 loc) · 11.7 KB
/
skipgram-trainer.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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
#!/eecs/research/asr/mingbin/python-workspace/hopeless/bin/python
"""
Author : Mingbin Xu (mingbin.xu@gmail.com)
Filename : skipgram-trainer.py
Last Update : Mar 25, 2016
Description : A wrapper of skip-gram from gensim
Website : https://wiki.eecs.yorku.ca/lab/MLL/
History
20160325 regex added to handle general number, date and phone number
Copyright (c) 2016 iNCML (author: Mingbin Xu)
License: MIT License (see ../LICENSE)
"""
import numpy
import gensim
import multiprocessing
import argparse
import os
import re
import os
import codecs
import logging
from hanziconv import HanziConv
logger = logging.getLogger()
__date1 = re.compile(
r"^(?:(?:31(\/|-|\.)(?:0?[13578]|1[02]|(?:Jan|Mar|May|Jul|Aug|Oct|Dec)))\1|(?:(?:29|30)(\/|-|\.)(?:0?[1,3-9]|1[0-2]|(?:Jan|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec))\2))(?:(?:1[6-9]|[2-9]\d)?\d{2})$|^(?:29(\/|-|\.)(?:0?2|(?:Feb))\3(?:(?:(?:1[6-9]|[2-9]\d)?(?:0[48]|[2468][048]|[13579][26])|(?:(?:16|[2468][048]|[3579][26])00))))$|^(?:0?[1-9]|1\d|2[0-8])(\/|-|\.)(?:(?:0?[1-9]|(?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep))|(?:1[0-2]|(?:Oct|Nov|Dec)))\4(?:(?:1[6-9]|[2-9]\d)?\d{2})$".encode('utf8')
)
__date2 = re.compile(
r"^(((\d{4}(\/|-|\.)((0[13578](\/|-|\.)|1[02](\/|-|\.))(0[1-9]|[12]\d|3[01])|(0[13456789](\/|-|\.)|1[012](\/|-|\.))(0[1-9]|[12]\d|30)|02(\/|-|\.)(0[1-9]|1\d|2[0-8])))|((([02468][048]|[13579][26])00|\d{2}([13579][26]|0[48]|[2468][048])))(\/|-|\.)02(\/|-|\.)29)){0,10}$".encode('utf8')
)
__number = re.compile(
r"^(\+|-)?(([1-9]\d{0,2}(,\d{3})*)|([1-9]\d*)|0)(\.\d+)?$".encode('utf8')
)
__phone = re.compile(
r"^(?:(?:\+?1\s*(?:[.-]\s*)?)?(?:\(\s*([2-9]1[02-9]|[2-9][02-8]1|[2-9][02-8][02-9])\s*\)|([2-9]1[02-9]|[2-9][02-8]1|[2-9][02-8][02-9]))\s*(?:[.-]\s*)?)?([2-9]1[02-9]|[2-9][02-9]1|[2-9][02-9]{2})\s*(?:[.-]\s*)?([0-9]{4})(?:\s*(?:#|x\.?|ext\.?|extension)\s*(\d+))?$".encode('utf8')
)
__time = re.compile(
r"^(?:(?:([01]?\d|2[0-3]):)?([0-5]?\d):)?([0-5]?\d)$".encode('utf8')
)
__has_digit_but_no_letter = re.compile( r"^(?=[^A-Za-z]+$).*[0-9].*$".encode('utf8') )
__has_digit = re.compile( r"^.*[0-9].*$".encode('utf8') )
def sentence_eng( filename, case_sensitive = False ):
"""
Parameters
----------
filename : str
directory or filename of the corpus, one sentence per line
Yields
------
sentence: list of str
a sentence splited as a list of words
"""
if os.path.isdir( filename ):
for f in os.listdir( filename ):
if os.path.isfile( filename + '/' + f ):
for s in sentence_eng( filename + '/' + f, case_sensitive ):
yield s
else:
logger.info( 'start to read ' + filename )
# with codecs.open( filename, 'rb', 'utf8' ) as fp:
with open( filename, 'rb' ) as fp:
for line in fp:
line = line.decode('utf-8','ignore').encode('utf-8')
if not case_sensitive:
s = line.strip().lower().split()
else:
s = line.strip().split()
for i in xrange(len(s)):
if re.match( __has_digit, s[i] ):
if re.match( __number, s[i] ):
s[i] = u'<numeric-value>'
elif re.match( __date1, s[i] ) or re.match( __date2, s[i] ):
s[i] = u'<date-value>'
elif re.match( __time, s[i] ):
s[i] = u'<time-value>'
elif re.match( __phone, s[i] ):
s[i] = u'<phone-value>'
# elif re.match( __digit_without_letter, s[i] ):
# s[i] = '<digit_without_letter>'
else:
s[i] = u'<contains-digit>'
yield s
def sentence_cmn( rspecifier, word_level = True ):
if os.path.isdir( rspecifier ):
for f in os.listdir( rspecifier ):
full_name = os.path.join( rspecifier, f )
for s in sentence_cmn( full_name, word_level ):
yield s
else:
with codecs.open( rspecifier, 'rb', 'utf8' ) as fp:
for line in fp:
line = HanziConv.toSimplified( line.strip() )
if word_level:
sent = line.split()
else:
sent = []
for w in line.split():
has_chinese = any( u'\u4e00' <= c <= u'\u9fff' for c in w )
if has_chinese:
sent.extend( list(w) )
else:
sent.append( w )
yield [ u'<numeric>' if re.match(__has_digit_but_no_letter, w) else w for w in sent ]
def skipgram_eng( filename, min_cnt, max_vocab, n_embedding, n_window,
word_list = None, case_sensitive = False ):
"""
Parameters
----------
filename : str
filename of the corpus, one sentence per line
min_cnt : int
minimum occurance of word to be considered
max_vocab : int
maximum number of words to keep
n_embedding : int
the dimension of word embedding
n_windows : int
the length of context
case_sensitive : bool
all words are transformed into lowercase if false
Returns:
word_vector : numpy.ndarray
the word vectors in a 2d matrix; positions are given by 'idx2word'
idx2word : list
the vocabulary in sorted order
"""
n_worker = multiprocessing.cpu_count()
logger.info( "This machine has %d processors. We'll use %d of them" %
(n_worker, n_worker) )
model = gensim.models.Word2Vec( min_count = min_cnt,
workers = n_worker,
size = n_embedding,
window = n_window,
max_vocab_size = max_vocab * 4 if max_vocab is not None else None,
sg = 1,
negative = 7 )
model.build_vocab( sentence_eng( filename, case_sensitive ) )
for _ in xrange( 10 ):
model.train( sentence_eng( filename, case_sensitive ) )
if os.path.isfile( 'questions-words.txt' ):
model.accuracy( 'questions-words.txt' )
if word_list is None:
idx2word = [ w for w in model.index2word if w != u'<unk>' and w != u'<UNK>' ]
if case_sensitive:
if max_vocab is not None:
idx2word = idx2word[:max_vocab - 2]
idx2word.append( u'<UNK>' )
if max_vocab is not None:
idx2word = idx2word[:max_vocab - 1]
idx2word.append( u'<unk>' )
else:
with codecs.open( word_list, 'rb', 'utf8' ) as fp:
idx2word = [ line.strip().split()[0] for line in fp ]
if not case_sensitive:
word_vector = numpy.ndarray( (len(idx2word), model.layer1_size), numpy.float32 )
for i in xrange( len(idx2word) - 1 ):
word_vector[i] = model[ idx2word[i] ]
word_vector[ len(idx2word) - 1 ] = word_vector[: len(idx2word) - 1].mean(0)
else:
n_lowercase_only = 0
word_vector = numpy.zeros( (len(idx2word), model.layer1_size), numpy.float32 )
for i in xrange( len(idx2word) - 2 ):
word_vector[i] = model[ idx2word[i] ]
if idx2word[i].islower() == idx2word[i]:
n_lowercase_only += 1
word_vector[-1] += word_vector[i]
else:
word_vector[-2] += word_vector[i]
if n_lowercase_only > 0:
word_vector[-1] = word_vector[-1] / n_lowercase_only
if len(idx2word) - 2 - n_lowercase_only > 0:
word_vector[-2] = word_vector[-2] / (len(idx2word) - 2 - n_lowercase_only)
#model.save( 'skipgram-last-run' )
return word_vector, idx2word
def skipgram_cmn( filename, min_cnt, max_vocab, n_embedding, n_window,
word_list = None, word_level = True ):
n_worker = multiprocessing.cpu_count()
logger.info( "This machine has %d processors. We'll use %d of them" %
(n_worker, n_worker) )
model = gensim.models.Word2Vec( min_count = min_cnt,
workers = n_worker,
size = n_embedding,
window = n_window,
max_vocab_size = max_vocab * 3 if max_vocab is not None else None,
sg = 1,
negative = 7 )
model.build_vocab( sentence_cmn( filename, word_level ) )
for _ in xrange( 7 ):
model.train( sentence_cmn( filename, word_level ) )
if word_list is None:
idx2word = [ w for w in model.index2word if w != u'<unk>' ]
if max_vocab is not None:
idx2word = idx2word[:max_vocab - 1]
idx2word.append( u'<unk>' )
else:
with codecs.open( word_list, 'rb', 'utf8' ) as fp:
idx2word = [ line.strip().split()[0] for line in fp ]
word_vector = numpy.ndarray( (len(idx2word), model.layer1_size), numpy.float32 )
for i in xrange( len(idx2word) - 1 ):
word_vector[i] = model[ idx2word[i] ]
word_vector[ len(idx2word) - 1 ] = word_vector[: len(idx2word) - 1].mean(0)
return word_vector, idx2word
if __name__ == '__main__':
logging.basicConfig( format = '%(asctime)s : %(levelname)s : %(message)s',
level= logging.INFO)
parser = argparse.ArgumentParser( description = 'a wrapper of SGNS' )
parser.add_argument( 'corpus', type = str, help = 'text file or a directory containing text files' )
parser.add_argument( 'basename', type = str, help = 'basename.{word2vec, wordlist} will be created' )
parser.add_argument( '--min_cnt', type = int, default = 10 )
parser.add_argument( '--max_vocab', type = int, default = 100000 )
parser.add_argument( '--n_word_embedding', type = int, default = 128 )
parser.add_argument( '--n_window', type = int, default = 7 )
parser.add_argument( '--word_list', type = str, default = None )
parser.add_argument( '--case_sensitive', action = 'store_true', default = False )
parser.add_argument( '--language', type = str, default = 'eng',
choices = [ 'eng', 'cmn', 'spa' ] )
parser.add_argument( '--word_level', action = 'store_true', default = False )
args = parser.parse_args()
logger.info( args )
if args.language == 'eng':
word2vec, idx2word = skipgram_eng( args.corpus,
args.min_cnt,
args.max_vocab,
args.n_word_embedding,
args.n_window,
args.word_list,
args.case_sensitive )
elif args.language == 'cmn':
word2vec, idx2word = skipgram_cmn( args.corpus,
args.min_cnt,
args.max_vocab,
args.n_word_embedding,
args.n_window,
args.word_list,
args.word_level )
with open( args.basename + '.word2vec', 'wb' ) as fp:
numpy.int32(word2vec.shape[0]).tofile( fp )
numpy.int32(word2vec.shape[1]).tofile( fp )
word2vec.tofile( fp )
with codecs.open( args.basename + '.wordlist', 'wb', 'utf8' ) as fp:
fp.write( u'\n'.join( idx2word ) )