-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_utils.py
226 lines (172 loc) · 7.92 KB
/
data_utils.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
# -*- coding:utf-8 -*-
""" Utilities for Data Operations.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import os
import sys
import sklearn
import numpy as np
import tensorflow as tf
pyVersion = sys.version_info[0]
class BaseDataLoader(object):
def __init__(self, batch_size):
self.train_inputs = None
self.train_targets = None
self.dropout_train_inputs = None
self.valid_inputs = None
self.valid_targets = None
self.test_inputs = None
self.test_targets = None
self.word_to_index = None
self.index_to_word = None
self.batch_size = batch_size
def train_generator(self):
for i in range(0, self.train_size, self.batch_size):
yield (self.train_inputs[i : i + self.batch_size],
self.train_targets[i : i + self.batch_size],
self.dropout_train_inputs[i : i + self.batch_size])
def valid_generator(self):
for i in range(0, self.valid_size, self.batch_size):
yield (self.valid_inputs[i : i + self.batch_size],
self.valid_targets[i : i + self.batch_size])
def test_generator(self):
for i in range(0, self.test_size, self.batch_size):
yield (self.test_inputs[i : i + self.batch_size],
self.test_targets[i : i + self.batch_size])
class IMDB(BaseDataLoader):
def __init__(self, batch_size, seq_len, word_dropout_rate, log_manager):
BaseDataLoader.__init__(self, batch_size)
self.seq_len = seq_len
self.word_dropout_rate = word_dropout_rate
self._index_from = 4
self.word_to_index, self.index_to_word = self._build_vocab()
self.UNK_ID = self.word_to_index["<unk>"]
self.EOS_ID = self.word_to_index["<eos>"]
self.PAD_ID = self.word_to_index["<pad>"]
self.SOS_ID = self.word_to_index["<sos>"]
log_manager.info("Vocabulary Loaded.")
self.train_inputs, self.train_targets, self.test_inputs, self.test_targets = self._load_data()
log_manager.info("IMDB Data Loaded.")
self.dropout_train_inputs = self.dropout_data()
log_manager.info("Train Inputs Dropout.")
self.train_size = len(self.train_inputs)
self.test_size = len(self.test_inputs)
self.vocab_size = len(self.word_to_index)
log_manager.info("IMDB Statistics:")
log_manager.info("Train Set Size: %d" % self.train_size)
log_manager.info("Test Set Size: %d" % self.test_size)
log_manager.info("Vocabulary Size: %d" % self.vocab_size)
def _build_vocab(self):
word_to_index = tf.contrib.keras.datasets.imdb.get_word_index()
word_to_index = {w: (i + self._index_from) for w,i in word_to_index.items()}
word_to_index["<pad>"] = 0
word_to_index["<sos>"] = 1
word_to_index["<unk>"] = 2
word_to_index["<eos>"] = 3
index_to_word = {i:w for (w,i) in word_to_index.items()}
index_to_word[-1] = "-1"
return word_to_index, index_to_word
def _load_data(self):
(train, _), (test, _) = tf.contrib.keras.datasets.imdb.load_data(num_words=None, index_from = self._index_from)
train_inputs, train_targets = self._pad(train)
test_inputs, test_targets = self._pad(test)
return train_inputs, train_targets, test_inputs, test_targets
def _pad(self, data):
inputs = []
targets = []
for ins in data:
if len(ins) < self.seq_len - 1:
inputs.append([self.SOS_ID] + ins + [self.PAD_ID] * (self.seq_len - 1 - len(ins)))
targets.append(ins + [self.EOS_ID] + [self.PAD_ID] * (self.seq_len - 1 - len(ins)))
else:
truncated = ins[:(self.seq_len - 1)]
inputs.append([self.SOS_ID] + truncated)
targets.append(truncated + [self.EOS_ID])
truncated = ins[-(self.seq_len-1):]
inputs.append([self.SOS_ID] + truncated)
targets.append(truncated + [self.EOS_ID])
return np.array(inputs), np.array(targets)
# if word_dropout_rate == 0, no words are dropped.
def _word_dropout(self, inputs):
is_dropped = np.random.binomial(1, self.word_dropout_rate, inputs.shape)
fn = np.vectorize(lambda inputs, k: self.UNK_ID if k else inputs)
return fn(inputs, is_dropped)
def dropout_data(self):
return self._word_dropout(self.train_inputs)
def shuffle(self):
self.train_inputs, self.train_targets, self.dropout_train_inputs = sklearn.utils.shuffle(self.train_inputs, self.train_targets, self.dropout_train_inputs)
self.test_inputs, self.test_targets = sklearn.utils.shuffle(self.test_inputs, self.test_targets)
class PTB(BaseDataLoader):
def __init__(self, batch_size, seq_len, word_dropout_rate, log_manager):
BaseDataLoader.__init__(self, batch_size)
_base_path = "data/ptb/simple-examples/data"
self.seq_len = seq_len
self.word_dropout_rate = word_dropout_rate
self.word_to_index, self.index_to_word = self._build_vocab(os.path.join(_base_path,"ptb.train.txt"))
self.UNK_ID = self.word_to_index["<unk>"]
self.EOS_ID = self.word_to_index["<eos>"]
self.PAD_ID = self.word_to_index["<pad>"]
self.SOS_ID = self.word_to_index["<sos>"]
log_manager.info("Vocabulary Loaded.")
self.train_inputs, self.train_targets = self._load_data(os.path.join(_base_path, "ptb.train.txt"))
self.valid_inputs, self.valid_targets = self._load_data(os.path.join(_base_path, "ptb.valid.txt"))
self.test_inputs, self.test_targets = self._load_data(os.path.join(_base_path, "ptb.test.txt"))
log_manager.info("PTB Data Loaded.")
self.dropout_train_inputs = self.dropout_data()
log_manager.info("Train Inputs Dropout.")
self.train_size = len(self.train_inputs)
self.valid_size = len(self.valid_inputs)
self.test_size = len(self.test_inputs)
self.vocab_size = len(self.word_to_index)
log_manager.info("PTB Statistics:")
log_manager.info("Train Set Size: %d" % self.train_size)
log_manager.info("Valid Set Size: %d" % self.valid_size)
log_manager.info("Test Set Size: %d" % self.test_size)
log_manager.info("Vocabulary Size: %d" % self.vocab_size)
def _read_words(self, filename):
with tf.gfile.GFile(filename, "r") as f:
if pyVersion == 3:
return f.read().replace("\n", "<eos>").split()
else:
return f.read().decode("utf-8").replace("\n", "<eos>").split()
def _build_vocab(self, filename):
data = self._read_words(filename)
counter = collections.Counter(data)
count_pairs = counter.most_common()
words, _ = list(zip(*count_pairs))
word_to_index = dict(zip(words, range(len(words))))
word_to_index = {w: (i + 2) for (w, i) in word_to_index.items()}
word_to_index["<pad>"] = 0
word_to_index["<sos>"] = 1
index_to_word = {i: w for (w, i) in word_to_index.items()}
index_to_word[-1] = "-1"
return word_to_index, index_to_word
def _load_data(self, filename):
words = self._read_words(filename)
raw_data = [self.word_to_index[word] for word in words if word in self.word_to_index]
data_len = len(raw_data)
batch_len = data_len // self.batch_size
data = np.reshape(raw_data[0:self.batch_size * batch_len], [self.batch_size, batch_len])
epoch_size = (batch_len - 1) // self.seq_len
inputs = []
targets = []
for i in range(epoch_size):
start_index = i * self.seq_len
end_index = (i + 1) * self.seq_len
inputs.append(data[:, start_index : end_index])
targets.append(data[:, start_index + 1 : end_index + 1])
return np.concatenate(inputs), np.concatenate(targets)
# if word_dropout_rate == 0, no words are dropped.
def _word_dropout(self, inputs):
is_dropped = np.random.binomial(1, self.word_dropout_rate, inputs.shape)
fn = np.vectorize(lambda inputs, k: self.UNK_ID if k else inputs)
return fn(inputs, is_dropped)
def dropout_data(self):
return self._word_dropout(self.train_inputs)
def shuffle(self):
self.train_inputs, self.train_targets, self.dropout_train_inputs = sklearn.utils.shuffle(self.train_inputs, self.train_targets, self.dropout_train_inputs)
self.valid_inputs, self.valid_targets = sklearn.utils.shuffle(self.valid_inputs, self.valid_targets)
self.test_inputs, self.test_targets = sklearn.utils.shuffle(self.test_inputs, self.test_targets)