-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathprepare_data.py
193 lines (177 loc) · 8.53 KB
/
prepare_data.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
# -*- coding: UTF-8 -*-
import numpy as np
import re
import os
import collections
from utils import plot_lengths
from config import CorpusType, TrainMode
class PrepareData:
def __init__(self, vocab_size, corpus, batch_length=224, batch_size=50, dict_path=None, mode=TrainMode.Batch,
type=CorpusType.Train):
self.vocab_size = vocab_size
self.dict_path = dict_path
self.batch_length = batch_length
self.batch_size = batch_size
self.SPLIT_CHAR = ' ' # 分隔符：双空格
# 字符数量，
# 其中'BATCH_PAD'表示构建batch时不足时补的字符，'UNK'表示词汇表外的字符，
# 'STAT'表示句子首字符之前的字符，'END'表示句子尾字符后面的字符，这两个字符用于生成字的上下文
self.count = [['BATCH_PAD', 0], ['UNK', 0], ['STRT', 0], ['END', 0]]
self.init_count = len(self.count)
if type == CorpusType.Train:
self.input_file = 'corpus/' + corpus + '_' + 'training.utf8'
self.output_base = 'corpus/' + corpus + '_' + 'training_'
elif type == CorpusType.Test:
self.input_file = 'corpus/' + corpus + '_' + 'test.utf8'
self.output_base = 'corpus/' + corpus + '_' + 'test_'
self.lines = self.read_lines()
if self.dict_path == None:
self.dictionary, self.reverse_dictionary = self.build_dictionary('corpus/' + corpus + '_dict.utf8')
else:
self.dictionary, self.reverse_dictionary = self.read_dictionary()
if self.dictionary == None:
print('vocabulary size larger than dictionary size')
exit(1)
if type == CorpusType.Train:
self.characters_index, self.labels_index = self.build_dataset()
if mode == TrainMode.Sentence:
self.characters_index, self.labels_index, self.lengths = self.build_batch(trunc=False)
np.save(self.output_base + 'characters', self.characters_index)
np.save(self.output_base + 'labels', self.labels_index)
np.save(self.output_base + 'lengths', self.lengths)
elif mode == TrainMode.Batch:
self.character_batches, self.label_batches, self.lengths, self.sentences, self.sentence_labels, self.sentence_lengths = self.build_batch()
np.save(self.output_base + 'character_batches', self.character_batches)
np.save(self.output_base + 'label_batches', self.label_batches)
np.save(self.output_base + 'lengths_batches', self.lengths)
elif type == CorpusType.Test:
self.raw_lines = list(map(lambda s: s.replace(self.SPLIT_CHAR, ''), self.lines))
if os.path.exists('corpus/' + corpus + '_test_labels.npy'):
self.labels_index = np.load('corpus/' + corpus + '_test_labels.npy')
else:
_, self.labels_index = self.build_dataset()
np.save('corpus/' + corpus + '_test_labels', self.labels_index)
# plot_lengths(self.sentence_lengths)
def read_lines(self):
file = open(self.input_file, 'r', encoding='utf-8')
content = file.read()
# sentences = re.sub('[ ]+', self.SPLIT_CHAR, strQ2B(content)).splitlines() # 将词分隔符统一为双空格
sentences = re.sub('[ ]+', self.SPLIT_CHAR, content).splitlines() # 将词分隔符统一为双空格
sentences = list(map(lambda s: s.strip(), filter(None, sentences))) # 去除空行，去首尾空格
file.close()
return sentences
def build_dictionary(self, output=None):
dictionary = {}
words = ''.join(self.lines).replace(' ', '')
vocab_count = len(collections.Counter(words))
print('characters count'+str(vocab_count))
if vocab_count + self.init_count < self.vocab_size:
return None
self.count.extend(collections.Counter(words).most_common(self.vocab_size - self.init_count))
for word, _ in self.count:
dictionary[word] = len(dictionary)
if output != None:
with open(output, 'w', encoding='utf8') as file:
for ch, index in dictionary.items():
file.write(ch + ' ' + str(index) + '\n')
reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
return dictionary, reverse_dictionary
def read_dictionary(self):
dict_file = open(self.dict_path, 'r', encoding='utf-8')
dict_content = dict_file.read().splitlines()
dictionary = {}
dict_arr = map(lambda item: item.split(' '), dict_content)
for _, dict_item in enumerate(dict_arr):
dictionary[dict_item[0]] = int(dict_item[1])
dict_file.close()
# if len(dictionary) < self.vocab_size:
# return None
reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
# else:
# reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys()))
# for i in range(self.vocab_size, len(dictionary)):
# dictionary.pop(reverse_dictionary[i])
return dictionary, reverse_dictionary
def build_dataset(self):
sentence_index = []
labels_index = []
for sentence in self.lines:
sentence_label = []
word_index = []
words = sentence.strip().split(self.SPLIT_CHAR)
for word in words:
l = len(word)
if l == 0:
continue
elif l == 1:
sentence_label.append(0)
else:
sentence_label.append(1)
sentence_label.extend([2] * (l - 2))
sentence_label.append(3)
for ch in word:
index = self.dictionary.get(ch)
if index is not None:
word_index.append(index)
else:
word_index.append(0)
sentence_index.append(word_index)
labels_index.append(sentence_label)
return np.array(sentence_index), np.array(labels_index)
def build_batch(self, trunc = True):
sentence_batches = []
label_batches = []
sentence_lengths = []
lengths = []
sentences = []
labels = []
unknown = 4
seg_ch = [self.dictionary['。'], self.dictionary['！'], self.dictionary['？']]
no_seg_ch = [self.dictionary['”']]
characters_index = self.characters_index.tolist()
labels_index = self.labels_index.tolist()
line_lengths = list(map(lambda chs: len(chs), characters_index))
def is_seg(item):
return item[1] in seg_ch and (item[0] < item[2] - 1 and characters[item[0] + 1] not in no_seg_ch)
for characters, label, length in zip(characters_index, labels_index, line_lengths):
if length <= 1:
continue
seg_indices = [0] + [i[0] + 1 for i in filter(is_seg, zip(range(length), characters, [length] * length))]
for pre_seg_index, cur_seg_index in zip(seg_indices[:-1], seg_indices[1:]):
sentence = characters[pre_seg_index:cur_seg_index]
sentence_labels = label[pre_seg_index:cur_seg_index]
sentences.append(sentence)
labels.append(sentence_labels)
sentence_length = len(sentence)
sentence_lengths.append(sentence_length)
if sentence_length <= self.batch_length:
pad_length = self.batch_length - sentence_length
sentence_batches.append(sentence + [self.dictionary['BATCH_PAD']] * pad_length)
label_batches.append(sentence_labels + [unknown] * pad_length)
lengths.append(sentence_length)
else:
if sentence_labels[sentence_length - 1] != 0 and 1 in sentence_labels[:sentence_length:-1]:
last_index = sentence_labels[:sentence_length:-1].index(1)
pad_length = self.batch_length - last_index
sentence_batches.append(sentence[:last_index] + [self.dictionary['BATCH_PAD']] * pad_length)
label_batches.append(sentence_labels[:last_index] + [unknown] * pad_length)
lengths.append(last_index)
else:
sentence_batches.append(sentence[:self.batch_length])
label_batches.append(sentence_labels[:self.batch_length])
lengths.append(self.batch_length)
if trunc:
extra_count = len(sentence_batches) % self.batch_size
sentence_batches = np.array(sentence_batches[:-extra_count], dtype=np.int32).reshape(
[-1, self.batch_size, self.batch_length])
label_batches = np.array(label_batches[:-extra_count], dtype=np.int32).reshape(
[-1, self.batch_size, self.batch_length])
lengths = np.array(lengths[:-extra_count], dtype=np.int32).reshape([-1, self.batch_size])
return sentence_batches, label_batches, lengths, sentences, labels, sentence_lengths
else:
return np.array(sentence_batches, dtype=np.int32),np.array(label_batches, dtype=np.int32), np.array(lengths,dtype=np.int32)
if __name__ == '__main__':
# PrepareData(4600, 'pku', mode=TrainMode.Batch)
# PrepareData(4000, 'pku', type=CorpusType.Test, dict_path='corpus/pku_dict.utf8')
# PrepareData(5000, 'msr', mode=TrainMode.Batch)
PrepareData(None, 'emr',mode=TrainMode.Sentence)