-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_batch.py
161 lines (127 loc) · 5.16 KB
/
data_batch.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
# -*- coding: utf-8 -*-
import os
import numpy as np
import config
import random
conf = config.config()
def create_data(data_dir, random_dir):
with open(random_dir, 'r') as f_random:
number=set(f_random.read().split())
f_random.close()
docs=[]
sums=[]
for each_file in os.listdir(data_dir):
if each_file[14:] in number:
with open(os.path.join(data_dir, each_file), 'r') as doc:
f=doc.read().split('\n\n')
docs.append(f[0].split())
sums.append(f[1].split())
return docs, sums
def vocab_build(train_doc, vocab_size):
data=[]
for doc in train_doc:
for word in doc:
data.append(word)
import collections
item_freq=collections.Counter(data).most_common(vocab_size)
vocab=[]
for word_freq in item_freq:
vocab.append(word_freq[0])
return vocab
def data_generation():
pub_dir = '/Users/zy/Desktop'
train_dir=pub_dir+'/training'
val_dir=pub_dir+'/val'
test_dir=pub_dir+'/test'
train_random=pub_dir+'/random_sample_for_train.txt'
val_random=pub_dir+'/random_sample_for_val.txt'
test_random=pub_dir+'/random_sample_for_test.txt'
train_doc, train_sum=create_data(train_dir, train_random)
val_doc, val_sum=create_data(val_dir, val_random)
test_doc, test_sum=create_data(test_dir, test_random)
vocab = vocab_build(train_doc, conf.vocab_size-4)
vocab.append('<unk>')
vocab.append('<go>')
vocab.append('<eos>')
vocab.append('<pad>')
return (train_doc, train_sum), (val_doc, val_sum), (test_doc, test_sum), vocab
def np_int(data):
return np.array(data).astype('int32')
def word2id(docu, summ, vocab, is_training=False):
vocab_set = set(vocab)
docu1=[]
docs_mask=[]
for d in docu:
d_temp = d[:conf.doc_len]
if is_training:
d_mask = np.array([1]*conf.doc_len).astype('int32')
d_mask[len(d_temp):] = 0
docs_mask.append(d_mask)
#pad if necessary
d_temp+= ['<pad>']*(conf.doc_len-len(d_temp))
d1 = []
for word in d_temp:
if word in vocab_set:
d1.append(vocab.index(word))
else:
d1.append(vocab.index('<unk>'))
docu1.append(d1)
if is_training:
y_true_labels=[]
deco_inputs=[]
summ_mask=[]
y_true1 = []
deco_input1 = []
for s in summ:
y_true = s[:conf.sum_len-1]+['<eos>']
deco_input = ['<go>']+s[:conf.sum_len-1]
s_mask = np.array([1]*conf.sum_len).astype('int32')
s_mask[len(y_true):] = 0
summ_mask.append(s_mask)
y_true+= ['<pad>']*(conf.sum_len-len(y_true))
deco_input+= ['<pad>']*(conf.sum_len-len(deco_input))
y_true1 = []
deco_input1 = []
for w1, w2 in zip(y_true, deco_input):
if w1 in vocab_set:
y_true1.append(vocab.index(w1))
else:
y_true1.append(vocab.index('<unk>'))
if w2 in vocab_set:
deco_input1.append(vocab.index(w2))
else:
deco_input1.append(vocab.index('<unk>'))
y_true_labels.append(y_true1)
deco_inputs.append(deco_input1)
return np_int(docu1), np_int(deco_inputs), np_int(y_true_labels), docs_mask, summ_mask
def batch_shuffle(train_doc2id, train_y_true2id, train_deco_inputs2id, doc_mask, sum_mask):
l = len(train_doc2id)
data_shuffle=[]
for i in range(l):
data_shuffle.append([train_doc2id[i], train_y_true2id[i], train_deco_inputs2id[i], doc_mask[i], sum_mask[i]])
random.shuffle(data_shuffle)
train_doc2id1=[]
train_y_true2id1=[]
train_deco_inputs2id1=[]
doc_mask1=[]
sum_mask1=[]
for each_tuple in data_shuffle:
train_doc2id1.append(each_tuple[0])
train_y_true2id1.append(each_tuple[1])
train_deco_inputs2id1.append(each_tuple[2])
doc_mask1.append(each_tuple[3])
sum_mask1.append(each_tuple[4])
return np_int(train_doc2id1), np_int(train_y_true2id1), np_int(train_deco_inputs2id1), np_int(doc_mask1), np.array(sum_mask1).astype("float32")
def data_batch():
print "data_batch..."
(train_doc, train_sum), (val_doc, val_sum), (test_doc, test_sum), vocab = data_generation()
print "finished!"
print "convert to id..."
train_doc2id, train_deco_inputs2id, train_y_true2id, doc_mask, sum_mask = word2id(train_doc, train_sum, vocab, is_training=True)
#val_doc2id, val_sum2id = word2id(val_doc, val_sum, vocab)
#test_doc2id, test_sum2id = word2id(test_doc, test_sum, vocab)
print "finished!"
print "shuffle..."
train_doc2id, train_y_true2id, train_deco_inputs2id, doc_mask, sum_mask = batch_shuffle(train_doc2id, train_y_true2id, train_deco_inputs2id, doc_mask, sum_mask)
print "finished!"
return train_doc2id, train_y_true2id, train_deco_inputs2id, doc_mask, sum_mask, vocab