-
Notifications
You must be signed in to change notification settings - Fork 23
/
datautils.py
257 lines (223 loc) · 8.35 KB
/
datautils.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
# coding=utf-8
import sys, nltk
import numpy as np
import mxnet as mx
import pickle
def Perplexity( label, pred ):
label = label.T.reshape((-1,))
loss = 0.
for i in range(pred.shape[0]):
loss += -np.log(max(1e-10, pred[i][int(label[i])]))
return np.exp(loss / label.size)
def default_read_content( path ):
with open(path) as ins:
content = ins.read()
return content
def default_build_vocab( vocab_path ):
vocab = default_read_content(vocab_path)
vocab = vocab.split('\n')
idx = 1
vocab_rsd = {}
for word in vocab:
vocab_rsd[word] = idx
idx += 1
return vocab, vocab_rsd
def default_gen_buckets( sentences, batch_size, the_vocab ):
len_dict = {}
max_len = -1
for sentence in sentences:
words = default_text2id(sentence, the_vocab)
if len(words) == 0:
continue
if len(words) > max_len:
max_len = len(words)
if len(words) in len_dict:
len_dict[len(words)] += 1
else:
len_dict[len(words)] = 1
# print(len_dict)
tl = 0
buckets = []
for l, n in len_dict.items():
if n + tl >= batch_size:
buckets.append(l)
tl = 0
else:
tl += n
if tl > 0:
buckets.append(max_len)
return buckets
def default_text2id( sentence, the_vocab, max_len, vocab ):
sentence = sentence.lower()
words = nltk.word_tokenize(sentence)
tokens = []
for w in words:
if len(w) == 0: continue
if len(tokens) >= max_len: break
if not w in vocab:
tokens.append(the_vocab['<unknown>'])
else:
tokens.append(the_vocab[w])
return tokens
def default_gen_buckets( len_dict, batch_size ):
tl = 0
buckets = []
for l, n in len_dict.items():
if n + tl >= batch_size:
buckets.append(l)
tl = 0
else:
tl += n
return buckets
class Seq2SeqIter(mx.io.DataIter):
def __init__( self, data_path, source_path, target_path, vocab, vocab_rsd, batch_size,
max_len, data_name='data', label_name='label', split_char='\n',
text2id=None, read_content=None, model_parallel=False, ctx=mx.cpu() ):
super(Seq2SeqIter, self).__init__()
self.ctx = ctx
self.iter_data = []
if data_path is not None:
print 'loading data set'
with open(data_path, 'r') as f:
self.iter_data = pickle.load(f)
self.size = len(self.iter_data)
self.vocab = vocab
self.vocab_rsd = vocab_rsd
self.vocab_size = len(vocab)
self.data_name = data_name
self.label_name = label_name
self.model_parallel = model_parallel
self.batch_size = batch_size
self.max_len = max_len
self.source_path = source_path
self.target_path = target_path
self.split_char = split_char
if text2id is None:
self.text2id = default_text2id
else:
self.text2id = text2id
if read_content is None:
self.read_content = default_read_content
else:
self.read_content = read_content
if len(self.iter_data) == 0:
self.iter_data = self.make_data_iter_plan()
def make_data_iter_plan( self ):
print 'processing the raw data '
source = self.read_content(self.source_path)
source_lines = source.split(self.split_char)
target = self.read_content(self.target_path)
target_lines = target.split(self.split_char)
self.size = len(source_lines)
self.suffer_ids = np.random.permutation(self.size)
self.enc_inputs = []
self.dec_inputs = []
self.dec_targets = []
len_dict = {}
cnt = 0
for i in range(self.size):
source = source_lines[i]
target = target_lines[i]
t1 = source.split('\t\t')
t2 = target.split('\t\t')
if len(t1) < 2 or len(t2) < 2: continue
st, su = t1[0], t1[1]
tt, tu = t2[0], t2[1]
dec_input = []
dec_target = []
s_tokens = self.text2id(st, self.vocab_rsd, self.max_len, self.vocab)
t_tokens = self.text2id(tt, self.vocab_rsd, self.max_len, self.vocab)
self.enc_inputs.append(s_tokens)
dec_input.append(self.vocab_rsd['<go>'])
dec_input[1:len(t_tokens) + 1] = t_tokens[:]
self.dec_inputs.append(dec_input)
dec_target[:len(t_tokens)] = t_tokens[:]
dec_target.append(self.vocab_rsd['<eos>'])
self.dec_targets.append(dec_target)
if len(dec_input) < 3: continue
if not len(dec_input) in len_dict.keys():
len_dict[len(dec_input)] = 1
else:
len_dict[len(dec_input)] += 1
cnt += 1
self.buckets = default_gen_buckets(len_dict, self.batch_size)
self.len_dict = len_dict
self.size = cnt
bucket_n_batches = {}
for l, n in self.len_dict.items():
if l < 3:
continue
bucket_n_batches[l] = n / self.batch_size
# print bucket_n_batches
data_buffer = {}
for i in range(self.size):
dec_input = self.dec_inputs[i]
if len(dec_input) < 3:
continue
enc_input = self.enc_inputs[i]
dec_target = self.dec_targets[i]
if not len(dec_input) in data_buffer.keys():
data_buffer[len(dec_input)] = []
data_buffer[len(dec_input)].append({
'enc_input': enc_input,
'dec_input': dec_input,
'dec_target': dec_target
})
else:
data_buffer[len(dec_input)].append({
'enc_input': enc_input,
'dec_input': dec_input,
'dec_target': dec_target
})
iter_data = []
for l, n in self.len_dict.items():
for k in range(0, n, self.batch_size):
if k + self.batch_size >= self.size: break
encin_batch = np.zeros((self.batch_size, self.max_len))
decin_batch = np.zeros((self.batch_size, l))
dectr_batch = np.zeros((self.batch_size, l))
if n < self.batch_size: break
for j in range(self.batch_size):
one = data_buffer[l][j]
encin = one['enc_input']
offset = self.max_len - len(encin)
encin_batch[j][offset:] = encin
decin_batch[j] = one['dec_input']
dectr_batch[j] = one['dec_target']
iter_data.append({
'enc_batch_in': encin_batch,
'dec_batch_in': decin_batch,
'dec_batch_tr': dectr_batch
})
with open('./data/data.pickle', 'w') as f:
print 'dumping data ...'
pickle.dump(iter_data, f)
return iter_data
def __iter__( self ):
for batch in self.iter_data:
yield batch
class SimpleBatch(object):
def __init__( self, data_names, data, label_names, label, bucket_key ):
self.data = data
self.label = label
self.data_names = data_names
self.label_names = label_names
self.bucket_key = bucket_key
self.pad = 0
self.index = None
@property
def provide_data( self ):
return [(n, x.shape) for n, x in zip(self.data_names, self.data)]
@property
def provide_label( self ):
return [(n, x.shape) for n, x in zip(self.label_names, self.label)]
if __name__ == '__main__':
vocab, vocab_rsd = default_build_vocab('./data/vocab.txt')
data = Seq2SeqIter(data_path=None, source_path='./data/a.txt', target_path='./data/b.txt',
vocab=vocab, vocab_rsd=vocab_rsd, batch_size=10, max_len=25,
data_name='data', label_name='label', split_char='\n',
text2id=None, read_content=None, model_parallel=False)
for iter in data:
print 'enc input size is (%d, %d), and dec size is (%d, %d)' % \
(iter['enc_batch_in'].shape[0], iter['enc_batch_in'].shape[1],
iter['dec_batch_in'].shape[0], iter['dec_batch_in'].shape[1])