-
Notifications
You must be signed in to change notification settings - Fork 6
/
train_manager.py
290 lines (243 loc) · 14.6 KB
/
train_manager.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
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
import tensorflow as tf
import numpy as np
import copy
from model import *
from input_data import ToyDataset
class TrainManager:
def __init__(self):
# set vocabulary dictionary
self.all_tokens = {}
self.all_tokens['<PAD>'] = 0
for ch in range(97, 119 + 1):
self.all_tokens[chr(ch)] = len(self.all_tokens)
self.all_tokens['<BOS>'] = len(self.all_tokens)
self.all_tokens['<EOS>'] = len(self.all_tokens)
self.all_tokens['<UNK>'] = len(self.all_tokens)
self.vocab_size = len(self.all_tokens)
# print(vocab_size)
# we treat x,y,z as OOV
self.vocab_dict = copy.deepcopy(self.all_tokens)
self.all_tokens['x'] = len(self.all_tokens)
self.all_tokens['y'] = len(self.all_tokens)
self.all_tokens['z'] = len(self.all_tokens)
self.reverse_vocab_dict = dict(zip(self.vocab_dict.values(), self.vocab_dict.keys()))
self.dataset = ToyDataset()
self.tested_examples = ['circumstances', 'affirmative', 'corresponding', 'caraphernology',
'experimentation', 'dizziness', 'harambelover', 'terrifyingly',
'axbycydxexfyzxxy']
def tok2idx(self, batch_data):
source_sent, target_sent = batch_data
batchsize = len(source_sent)
encoder_max_length = 0
for sent in source_sent:
encoder_max_length = max(encoder_max_length, len(sent))
encoder_inputs = np.zeros(dtype=int, shape=(batchsize, encoder_max_length))
for i, sent in enumerate(source_sent):
for j, word in enumerate(sent):
if word in self.vocab_dict:
encoder_inputs[i][j] = self.vocab_dict[word]
else:
encoder_inputs[i][j] = self.vocab_dict['<UNK>']
decoder_max_length = 0
for sent in target_sent:
decoder_max_length = max(decoder_max_length, len(sent))
decoder_inputs = np.zeros(dtype=int, shape=(batchsize, decoder_max_length + 1))
decoder_outputs = np.zeros(dtype=int, shape=(batchsize, decoder_max_length + 1))
for i, sent in enumerate(target_sent):
for j, word in enumerate(sent):
if word in self.vocab_dict:
decoder_inputs[i][j + 1] = self.vocab_dict[word]
decoder_outputs[i][j] = self.vocab_dict[word]
else:
decoder_inputs[i][j + 1] = self.vocab_dict['<UNK>']
decoder_outputs[i][j] = self.vocab_dict['<UNK>']
decoder_inputs[i][0] = self.vocab_dict['<BOS>']
decoder_outputs[i][len(sent)] = self.vocab_dict['<EOS>']
return encoder_inputs, decoder_inputs, decoder_outputs
def run_model(self):
pass
class Train_BasicEncoderDecoder(TrainManager):
def run_model(self):
model = BasicEncoderDecoder(name='BasicEncoderDecoder', vocab_size=self.vocab_size)
plot_every_steps = 100
print('train BasicEncoderDecoder ... ')
with tf.Session() as sess:
sess.run(tf.group(tf.global_variables_initializer()))
average_loss = 0
for step in range(1, 50000):
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx(self.dataset.get_batch())
_, training_loss = sess.run([model.train_op, model.train_loss],
feed_dict={
model.encoder_inputs:encoder_inputs,
model.decoder_inputs:decoder_inputs,
model.decoder_outputs:decoder_outputs,
model.keep_prob: 0.5})
average_loss += training_loss / plot_every_steps
if step % plot_every_steps == 0:
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx(self.dataset.get_test_data())
accu = sess.run(model.accuracy,
feed_dict={
model.encoder_inputs:encoder_inputs,
model.decoder_inputs:decoder_inputs,
model.decoder_outputs:decoder_outputs,
model.keep_prob: 1.0})
# test some difficult examples
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx((self.tested_examples, self.tested_examples))
pred = sess.run(model.predict,
feed_dict={
model.encoder_inputs: encoder_inputs,
model.decoder_inputs: decoder_inputs,
model.decoder_outputs: decoder_outputs,
model.keep_prob: 1.0})
for ii in range(len(self.tested_examples)):
print()
print('true output:',
" ".join([self.reverse_vocab_dict[word_id] for word_id in decoder_outputs[ii]]))
print('pred output:', " ".join([self.reverse_vocab_dict[word_id] for word_id in pred[ii]]))
print("step %5d, loss=%0.4f accu=%0.4f" % (step, average_loss, accu))
average_loss = 0
class Train_AttenNet(TrainManager):
def run_model(self):
model = AttenNet(name='AttenNet', vocab_size=self.vocab_size)
plot_every_steps = 100
print('train EncoderDecoder with attention mechanism ... ')
with tf.Session() as sess:
sess.run(tf.group(tf.global_variables_initializer()))
average_loss = 0
for step in range(1, 50000):
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx(self.dataset.get_batch())
_, training_loss = sess.run([model.train_op, model.train_loss],
feed_dict={
model.encoder_inputs:encoder_inputs,
model.decoder_inputs:decoder_inputs,
model.decoder_outputs:decoder_outputs,
model.keep_prob: 0.5})
average_loss += training_loss / plot_every_steps
if step % plot_every_steps == 0:
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx(self.dataset.get_test_data())
accu = sess.run(model.accuracy,
feed_dict={
model.encoder_inputs:encoder_inputs,
model.decoder_inputs:decoder_inputs,
model.decoder_outputs:decoder_outputs,
model.keep_prob: 1.0})
# test some difficult examples
encoder_inputs, decoder_inputs, decoder_outputs, \
= self.tok2idx((self.tested_examples, self.tested_examples))
pred, cum_att_weights = sess.run([model.predict, model.cum_att_weights],
feed_dict={
model.encoder_inputs: encoder_inputs,
model.decoder_inputs: decoder_inputs,
model.decoder_outputs: decoder_outputs,
model.keep_prob: 1.0})
for ii in range(len(self.tested_examples)):
print()
print('true output:',
" ".join([self.reverse_vocab_dict[word_id] for word_id in decoder_outputs[ii]]))
print('pred output:', " ".join([self.reverse_vocab_dict[word_id] for word_id in pred[ii]]))
print(' ', end=' ')
for word_id in encoder_inputs[ii]:
print('%5s' % self.reverse_vocab_dict[word_id], end=' ')
print()
for i, prob in enumerate(cum_att_weights[ii]):
print('%5s' % self.reverse_vocab_dict[decoder_outputs[ii][i]], end=' ')
for p in prob:
print('%0.3f' % p, end=' ')
else:
print()
print()
print("step %5d, loss=%0.4f accu=%0.4f" % (step, average_loss, accu))
average_loss = 0
class Train_CopyNet(TrainManager):
def tok2idx(self, batch_data):
source_sent, target_sent = batch_data
batchsize = len(source_sent)
encoder_max_length = 0
for sent in source_sent:
encoder_max_length = max(encoder_max_length, len(sent))
encoder_inputs = np.zeros(dtype=int, shape=(batchsize, encoder_max_length))
batch_OOV_tokens = []
for i, sent in enumerate(source_sent):
OOV_token = list(set([w for w in sent if w not in self.vocab_dict]))
batch_OOV_tokens.append(OOV_token)
for j, token in enumerate(sent):
if token in self.vocab_dict:
encoder_inputs[i][j] = self.vocab_dict[token]
else:
encoder_inputs[i][j] = self.vocab_size + OOV_token.index(token)
decoder_max_length = 0
for sent in target_sent:
decoder_max_length = max(decoder_max_length, len(sent))
decoder_inputs = np.zeros(dtype=int, shape=(batchsize, decoder_max_length + 1))
decoder_outputs = np.zeros(dtype=int, shape=(batchsize, decoder_max_length + 1))
for i, sent in enumerate(target_sent):
for j, token in enumerate(sent):
if token in self.vocab_dict:
decoder_inputs[i][j + 1] = self.vocab_dict[token]
decoder_outputs[i][j] = self.vocab_dict[token]
elif token in batch_OOV_tokens[i]:
decoder_inputs[i][j + 1] = self.vocab_size + batch_OOV_tokens[i].index(token)
decoder_outputs[i][j] = self.vocab_size + batch_OOV_tokens[i].index(token)
else:
decoder_inputs[i][j + 1] = self.vocab_dict['<UNK>']
decoder_outputs[i][j] = self.vocab_dict['<UNK>']
decoder_inputs[i][0] = self.vocab_dict['<BOS>']
decoder_outputs[i][len(sent)] = self.vocab_dict['<EOS>']
return encoder_inputs, decoder_inputs, decoder_outputs, batch_OOV_tokens
def run_model(self):
model = CopyNet(name='CopyNet', vocab_size=self.vocab_size)
plot_every_steps = 100
print('train EncoderDecoder with copy mechanism ... ')
with tf.Session() as sess:
sess.run(tf.group(tf.global_variables_initializer()))
average_loss = 0
for step in range(1, 50000):
encoder_inputs, decoder_inputs, decoder_outputs, batch_OOV_tokens \
= self.tok2idx(self.dataset.get_batch())
_, training_loss = sess.run([model.train_op, model.train_loss],
feed_dict={model.encoder_inputs: encoder_inputs,
model.decoder_inputs: decoder_inputs,
model.decoder_outputs: decoder_outputs,
model.batch_OOV_num: np.max([len(tokens) for tokens in
batch_OOV_tokens]),
model.keep_prob: 0.5})
average_loss += training_loss / plot_every_steps
if step % plot_every_steps == 0:
encoder_inputs, decoder_inputs, decoder_outputs, batch_OOV_tokens \
= self.tok2idx(self.dataset.get_test_data())
accu = sess.run(model.accuracy,
feed_dict={model.encoder_inputs: encoder_inputs,
model.decoder_inputs: decoder_inputs,
model.decoder_outputs: decoder_outputs,
model.batch_OOV_num: np.max([len(tokens) for tokens in
batch_OOV_tokens]),
model.keep_prob: 1.0})
encoder_inputs, decoder_inputs, decoder_outputs, batch_OOV_tokens \
= self.tok2idx((self.tested_examples, self.tested_examples))
pred = sess.run(model.predict,
feed_dict={model.encoder_inputs: encoder_inputs,
model.decoder_inputs: decoder_inputs,
model.decoder_outputs: decoder_outputs,
model.batch_OOV_num: np.max([len(tokens) for tokens in
batch_OOV_tokens]),
model.keep_prob: 1.0})
for ii in range(len(self.tested_examples)):
print()
print('true output:', " ".join([self.reverse_vocab_dict[word_id]
if word_id < self.vocab_size
else batch_OOV_tokens[ii][word_id - self.vocab_size]
for word_id in decoder_outputs[ii]]
))
print('pred output:', " ".join([self.reverse_vocab_dict[word_id]
if word_id < self.vocab_size
else batch_OOV_tokens[ii][word_id - self.vocab_size]
for word_id in pred[ii]]
))
print()
print("step %5d, loss=%0.4f accu=%0.4f" % (step, average_loss, accu))
average_loss = 0