/
subtoken_vocab.py
236 lines (201 loc) · 8.02 KB
/
subtoken_vocab.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
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
# Copyright 2016 Timothy Dozat
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import codecs
from collections import Counter
import numpy as np
import tensorflow as tf
from parser.vocabs import TokenVocab
from parser import Multibucket
from parser.misc.bucketer import Bucketer
__all__ = ['CharVocab']
#***************************************************************
class SubtokenVocab(TokenVocab):
""""""
#=============================================================
def __init__(self, token_vocab, *args, **kwargs):
""""""
recount = kwargs.pop('recount', False)
initialize_zero = kwargs.pop('initialize_zero', False)
super(TokenVocab, self).__init__(*args, **kwargs)
self._token_vocab = token_vocab
self._token_counts = Counter()
self._multibucket = Multibucket.from_configurable(self, embed_model=self.embed_model, name=self.name)
self._tok2idx = {}
if recount:
self.count()
else:
if os.path.isfile(self.filename):
self.load()
else:
self.count()
self.dump()
self.index_vocab()
embed_dims = [len(self), self.embed_size]
if initialize_zero:
self._embeddings_array = np.zeros(embed_dims)
else:
self._embeddings_array = np.random.randn(*embed_dims)
return
#=============================================================
def setup(self):
""""""
self.placeholder = None
with tf.device('/cpu:0'):
with tf.variable_scope(self.name.title()):
self._embeddings = tf.Variable(self._embeddings_array, name='Embeddings', dtype=tf.float32, trainable=True)
self._multibucket.reset_placeholders()
return
#=============================================================
def __call__(self, placeholder=None, moving_params=None):
""""""
placeholder = self.generate_placeholder() if placeholder is None else placeholder
embeddings = self.multibucket(self, keep_prob=self.embed_keep_prob, moving_params=moving_params)
return tf.nn.embedding_lookup(embeddings, placeholder)
#=============================================================
def count(self):
""""""
special_tokens = set(self.token_vocab.special_tokens)
for token in self.token_vocab.counts:
for subtoken in token:
self.counts[subtoken] += 1
self.token_counts[subtoken] += self.token_vocab.counts[token]
return
#=============================================================
def load(self):
""""""
train_file = os.path.join(self.save_dir, self.name+'.txt')
with codecs.open(train_file, encoding='utf-8') as f:
for line_num, line in enumerate(f):
try:
line = line.rstrip()
if line:
line = line.split('\t')
token, count, token_count = line
self.counts[token] = int(count)
self.token_counts[token] = int(token_count)
except:
raise ValueError('File %s is misformatted at line %d' % (train_file, line_num+1))
return
#=============================================================
def dump(self):
""""""
with codecs.open(os.path.join(self.save_dir, self.name+'.txt'), 'w', encoding='utf-8') as f:
for token, count in self.sorted_counts(self._counts):
f.write('%s\t%d\t%d\n' % (token, count, self.token_counts[token]))
return
#=============================================================
def subtoken_indices(self, token):
""""""
return self[list(token)]
#=============================================================
def index_tokens(self):
""""""
self._tok2idx = {}
tok2idxs = {token: self.subtoken_indices(token) for token in self.token_vocab.counts}
with Bucketer.from_configurable(self, self.n_buckets, name='bucketer-%s'%self.name) as bucketer:
splits = bucketer.compute_splits(len(indices) for indices in tok2idxs.values())
with self.multibucket.open(splits):
for index, special_token in enumerate(self.token_vocab.special_tokens):
index = index if index != self.token_vocab.UNK else self.META_UNK
self.tok2idx[special_token] = self.multibucket.add([index])
for token, _ in self.sorted_counts(self.token_vocab.counts):
self.tok2idx[token] = self.multibucket.add(tok2idxs[token])
self._idx2tok = {idx: tok for tok, idx in self.tok2idx.iteritems()}
self._idx2tok[0] = self[self.PAD]
return
#=============================================================
def set_feed_dict(self, data, feed_dict):
""""""
uniq, inv = np.unique(data, return_inverse=True)
# this placeholder stores the indices into the new, on-the-fly embedding matrix
feed_dict[self.placeholder] = inv.reshape(data.shape)
unsorted = []
indices = self.multibucket.indices[uniq]
for bkt_idx, bucket in enumerate(self.multibucket):
where = np.where(indices['bkt_idx'] == bkt_idx)[0]
idxs = indices[where]['idx']
bucket_data = bucket.indices[idxs]
# these placeholders store the bucket data's index into the vocab's subtoken matrix
if bucket_data.shape[0]:
unsorted.append(where)
feed_dict[bucket.placeholder] = bucket_data
else:
feed_dict[bucket.placeholder] = bucket.indices[0:1]
# this placeholder makes sure the on-the-fly embedding matrix is in the right order
feed_dict[self.multibucket.placeholder] = np.argsort(np.concatenate(unsorted))
return
#=============================================================
def index(self, token):
if not self.cased and token not in self._special_tokens_set:
token = token.lower()
return self._tok2idx.get(token, self.META_UNK)
#=============================================================
@property
def multibucket(self):
return self._multibucket
@property
def token_counts(self):
return self._token_counts
@property
def token_vocab(self):
return self._token_vocab
@property
def token_embed_size(self):
return (self.token_vocab or self).embed_size
@property
def conll_idx(self):
return self.token_vocab.conll_idx
@property
def tok2idx(self):
return self._tok2idx
@property
def idx2tok(self):
return self._idx2tok
#=============================================================
def __setattr__(self, name, value):
if name == '_token_vocab':
if self.cased is None:
self._cased = value.cased
elif self.cased != value.cased:
cls = value.__class__
value = cls.from_configurable(value,
cased=self.cased,
recount=True)
super(SubtokenVocab, self).__setattr__(name, value)
return
#***************************************************************
class CharVocab(SubtokenVocab):
pass
#***************************************************************
if __name__ == '__main__':
""""""
from parser import Configurable
from parser.vocabs import WordVocab, CharVocab
configurable = Configurable()
token_vocab = WordVocab.from_configurable(configurable, 1)
token_vocab.fit_to_zipf()
if os.path.isfile('saves/defaults/chars.txt'):
os.remove('saves/defaults/chars.txt')
subtoken_vocab = CharVocab.from_vocab(token_vocab)
subtoken_vocab = CharVocab.from_vocab(token_vocab)
subtoken_vocab.token_vocab.count(configurable.valid_files)
subtoken_vocab.index_tokens()
subtoken_vocab.fit_to_zipf()
print('SubtokenVocab passes')