/
data_helper.py
218 lines (172 loc) · 8.52 KB
/
data_helper.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
import numpy as np
import os
import pickle
import codecs
import collections
from keras.utils.np_utils import to_categorical
class Vocab:
def __init__(self, token2index=None, index2token=None):
self._token2index = token2index or {}
self._index2token = index2token or []
def feed(self, token):
if token not in self._token2index:
# allocate new index for this token
index = len(self._token2index)
self._token2index[token] = index
self._index2token.append(token)
return self._token2index[token]
@property
def size(self):
return len(self._token2index)
def token(self, index):
return self._index2token[index]
def __getitem__(self, token):
index = self.get(token)
if index is None:
raise KeyError(token)
return index
def get(self, token, default=None):
return self._token2index.get(token, default)
@classmethod
def load(cls, filename):
with open(filename, 'rb') as f:
token2index, index2token = pickle.load(f)
return cls(token2index, index2token)
def load_data(file_name,max_sentnece_length,max_word_length):
char_vocab = Vocab()
char_vocab.feed(' ')
word_vocab = Vocab()
word_vocab.feed(' ')
label_vocab = Vocab()
label_vocab.feed('+')
word_tokens = collections.defaultdict(list)
char_tokens = collections.defaultdict(list)
label_tokens = collections.defaultdict(list)
input_word = collections.defaultdict(list)
input_char = collections.defaultdict(list)
input_label = collections.defaultdict(list)
mask_label = collections.defaultdict(list)
for data_name in ('train', 'test', 'dev', 'domain_ptb'):
print('reading data:', data_name)
file = codecs.open(os.path.join(file_name,data_name), 'r', 'utf-8')
# onesplit = []
# labelsplit = []
sentence = []#each sentence padding to max length
word_char = []#padding word conrespanding to ' '
sen_label = []#each sentence label padding to max length
mask = []
a=0
for line in file:
line = line.strip()
if line!='':
onesplit=line.split('\t')[0]
labelsplit=line.split('\t')[1]
# for word,label in zip(onesplit,labelsplit):
word_tokens[data_name].append(word_vocab.feed(onesplit))
mask.append(1)
if len(onesplit) > max_word_length:
char_array = [char_vocab.feed(c) for c in onesplit[:max_word_length]]
else:
char_array = [char_vocab.feed(c) for c in onesplit]
char_tokens[data_name].append(char_array)
label_tokens[data_name].append(label_vocab.feed(labelsplit))
sentence.append(word_vocab.feed(onesplit))
word_char.append(char_array)
sen_label.append(label_vocab.feed(labelsplit))
else:
a+=1
if len(sentence) < max_sentnece_length:#padding
for i in range(max_sentnece_length - len(sentence)):
sentence.append(word_vocab.feed(' '))
mask.append(0)
sen_label.append(label_vocab.feed('+'))
word_char.append([char_vocab.feed(' ')])
if len(sentence) > max_sentnece_length:#cutting
sentence = sentence[:max_sentnece_length]
mask = mask[:max_sentnece_length]
sen_label = sen_label[:max_sentnece_length]
word_char = word_char[:max_sentnece_length]
input_word[data_name].append(sentence)
input_char[data_name].append(word_char)
input_label[data_name].append(sen_label)
mask_label[data_name].append(mask)
if len(sentence)!=max_sentnece_length:
print(sentence)
sentence = []
sen_label = []
word_char = []
mask = []
# onesplit = []
# labelsplit = []
##############################################################################
word_tensors = {}
char_tensors = {}
label_tensors = {}
mask_tensors = {}
for fname in ('train', 'test', 'dev', 'domain_ptb',):
assert len(input_word[fname]) == len(input_char[fname])
assert len(input_word[fname]) == len(input_label[fname])
print('tranforming numpy array:',fname)
# for sentence in input_word[fname]:
# try:
# word_tensors[fname] = np.append(word_tensors[fname],np.array(sentence, dtype=np.int32),axis=0)
# except ValueError:
# file=open('testtxt','w')
# file.write(str(sentence))
# assert 1==2
word_tensors[fname] = np.array(input_word[fname],dtype=np.int32)
char_tensors[fname] = np.zeros([len(input_char[fname]), max_sentnece_length, max_word_length], dtype=np.int32)
label_tensors[fname] = np.array(input_label[fname], dtype=np.int32)
mask_tensors[fname] = np.array(mask_label[fname], dtype=np.int32)
for i, word_array in enumerate(input_char[fname]):
for j,char_array in enumerate(word_array):
char_tensors[fname][i,j, :len(char_array)] = np.array(char_array,dtype=np.int32)
print('label vocab size:',label_vocab.size)
print('label vocab index to token:',label_vocab._index2token)
return word_vocab, char_vocab, label_vocab, word_tensors, char_tensors, label_tensors,mask_tensors
#word_tensors:[num_sentence, max_sentence_length]
#char_tensors:[num_sentence, max_sentence_length, max_word_length]
#label_tensors:[num_sentence, max_sentence_length]
class DataReader:
def __init__(self, word_tensor, char_tensor,label_tensor, mask_tensor, batch_size, num_class):
length = word_tensor.shape[0]
assert char_tensor.shape[0] == length
max_sentence_length = char_tensor.shape[1]
max_word_length = char_tensor.shape[2]
#for better reshape
reduced_length = (length // batch_size) * batch_size
word_tensor1 = word_tensor[:reduced_length,:]
self.word_tensor2 = word_tensor[reduced_length:, :]#not use, because use batch_size =1 when test
char_tensor1 = char_tensor[:reduced_length, :,:]
self.char_tensor2 = char_tensor[reduced_length:, :, :]
mask_tensor1 = mask_tensor[:reduced_length, :]
self.mask_tensor2 = mask_tensor[reduced_length:, :]
label_tensor = np.reshape(label_tensor,[-1])
label_tensor = to_categorical(label_tensor,num_classes=num_class)
label_tensor = np.reshape(label_tensor,[length,max_sentence_length,num_class])
label_tensor1 = label_tensor[:reduced_length,:,:]
self.label_tensor2 = label_tensor[reduced_length:,:,:]
w_batches = word_tensor1.reshape([-1, batch_size,max_sentence_length])
x_batches = char_tensor1.reshape([-1, batch_size, max_sentence_length, max_word_length])
y_batches = label_tensor1.reshape([-1, batch_size, max_sentence_length,num_class])
m_batches = mask_tensor1.reshape([-1, batch_size,max_sentence_length])
self._w_batches = list(w_batches)
self._x_batches = list(x_batches)
self._y_batches = list(y_batches)
self._m_batches = list(m_batches)
assert len(self._x_batches) == len(self._y_batches)==len(self._w_batches)==len(self._m_batches)
self.length = len(self._y_batches)
self.batch_size = batch_size
self.num_unroll_steps = max_sentence_length
def iter(self):
for x, y, z, m in zip(self._x_batches, self._y_batches, self._w_batches, self._m_batches):
yield x, y, z, m
if __name__ == '__main__':
word_vocab, char_vocab, label_vocab, word_tensors, char_tensors, label_tensors,mask_tensors = load_data(file_name='../',max_sentnece_length=39,max_word_length=35)
a = DataReader(word_tensors['domain_ptb'], char_tensors['domain_ptb'],label_tensors['domain_ptb'],mask_tensors['domain_ptb'], 20, 54)
for c,d,e,m in a.iter():
print('cccccccccccccccccccccccccccc',c)
print('dddddddddddddddddddddddddddd',d)
print('eeeeeeeeeeeeeeeeeeeeeeeeeeee',e)
print('mmmmmmmmmmmmmmmmmmmmmmmmmmmm',m)
assert 1==2