-
-
Notifications
You must be signed in to change notification settings - Fork 91
/
seqgan.py
398 lines (305 loc) · 14.1 KB
/
seqgan.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
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
import argparse
import re
from typing import Dict, List, Tuple, Set
import numpy as np
import torch
import torch.optim as optim
from allennlp.common.util import START_SYMBOL, END_SYMBOL
from allennlp.data.fields import LabelField, TextField
from allennlp.data.instance import Instance
from allennlp.data.iterators import BasicIterator
from allennlp.data.token_indexers import TokenIndexer, SingleIdTokenIndexer
from allennlp.data.tokenizers.character_tokenizer import CharacterTokenizer
from allennlp.data.tokenizers.token import Token
from allennlp.data.vocabulary import Vocabulary, DEFAULT_PADDING_TOKEN
from allennlp.models import Model
from allennlp.modules.seq2seq_encoders import PytorchSeq2SeqWrapper
from allennlp.modules.seq2vec_encoders import CnnEncoder
from allennlp.modules.text_field_embedders import TextFieldEmbedder, BasicTextFieldEmbedder
from allennlp.modules.token_embedders import Embedding
from allennlp.nn.util import get_text_field_mask, sequence_cross_entropy_with_logits
from allennlp.training.trainer import Trainer
from nltk.translate.chrf_score import sentence_chrf
class Generator(Model):
def __init__(self,
embedder: TextFieldEmbedder,
embedding_size: int,
hidden_size: int,
max_len: int,
vocab: Vocabulary) -> None:
super().__init__(vocab)
self.embedder = embedder
self.rnn = PytorchSeq2SeqWrapper(
torch.nn.LSTM(embedding_size, hidden_size, num_layers=1, batch_first=True))
self.hidden2out = torch.nn.Linear(in_features=self.rnn.get_output_dim(),
out_features=vocab.get_vocab_size('tokens'))
self.hidden_size = hidden_size
self.max_len = max_len
def forward(self, input_tokens, output_tokens):
mask = get_text_field_mask(input_tokens)
embeddings = self.embedder(input_tokens)
rnn_hidden = self.rnn(embeddings, mask)
out_logits = self.hidden2out(rnn_hidden)
loss = sequence_cross_entropy_with_logits(out_logits, output_tokens['tokens'], mask)
return {'loss': loss}
def generate(self) -> Tuple[List[Token], torch.tensor]:
start_symbol_idx = self.vocab.get_token_index(START_SYMBOL, 'tokens')
end_symbol_idx = self.vocab.get_token_index(END_SYMBOL, 'tokens')
padding_symbol_idx = self.vocab.get_token_index(DEFAULT_PADDING_TOKEN, 'tokens')
log_likelihood = 0.
words = []
state = (torch.zeros(1, 1, self.hidden_size),
torch.zeros(1, 1, self.hidden_size))
word_idx = start_symbol_idx
for i in range(self.max_len):
tokens = torch.tensor([[word_idx]])
embeddings = self.embedder({'tokens': tokens})
output, state = self.rnn._module(embeddings, state)
output = self.hidden2out(output)
log_prob = torch.log_softmax(output[0, 0], dim=0)
dist = torch.exp(log_prob)
word_idx = start_symbol_idx
while word_idx in {start_symbol_idx, padding_symbol_idx}:
word_idx = torch.multinomial(
dist, num_samples=1, replacement=False).item()
log_likelihood += log_prob[word_idx]
if word_idx == end_symbol_idx:
break
token = Token(text=self.vocab.get_token_from_index(word_idx, 'tokens'))
words.append(token)
return words, log_likelihood
class Discriminator(Model):
def __init__(self,
embedder: TextFieldEmbedder,
embedding_size: int,
num_filters: int,
vocab: Vocabulary) -> None:
super().__init__(vocab)
self.embedder = embedder
self.encoder = CnnEncoder(embedding_size, num_filters=num_filters)
self.linear = torch.nn.Linear(in_features=self.encoder.get_output_dim(),
out_features=vocab.get_vocab_size('labels'))
self.loss_function = torch.nn.CrossEntropyLoss()
def forward(self,
tokens: Dict[str, torch.Tensor],
label: torch.Tensor = None) -> Dict[str, torch.Tensor]:
mask = get_text_field_mask(tokens)
embeddings = self.embedder(tokens)
encoder_out = self.encoder(embeddings, mask)
logits = self.linear(encoder_out)
output = {"logits": logits}
output["loss"] = self.loss_function(logits, label)
return output
def read_shakespeare(all_chars: Set[str]=None) -> List[List[Token]]:
tokenizer = CharacterTokenizer()
sentences = []
with open('data/shakespeare/hamlet.txt') as f:
for line in f:
line = line.strip()
if not line:
continue
line = re.sub(' +', ' ', line)
tokens = tokenizer.tokenize(line)
if all_chars:
tokens = [token for token in tokens if token.text in all_chars]
sentences.append(tokens)
return sentences
def text_to_disc_instance(tokens: List[Token],
label: str,
token_indexers: Dict[str, TokenIndexer]):
fields = {'tokens': TextField(tokens, token_indexers),
'label': LabelField(label)}
return Instance(fields)
def tokens_to_lm_instance(tokens: List[Token],
token_indexers: Dict[str, TokenIndexer]):
tokens = list(tokens) # shallow copy
tokens.insert(0, Token(START_SYMBOL))
tokens.append(Token(END_SYMBOL))
input_field = TextField(tokens[:-1], token_indexers)
output_field = TextField(tokens[1:], token_indexers)
return Instance({'input_tokens': input_field,
'output_tokens': output_field})
def get_discriminator_batch(generator: Generator,
train_set: List[List[Token]],
token_indexers: Dict[str, TokenIndexer],
batch_size: int) -> List[Instance]:
# Generate real batch
instances = []
num_samples = min(len(train_set), batch_size)
sent_ids = np.random.choice(len(train_set), size=num_samples, replace=False)
for sent_id in sent_ids:
tokens = train_set[sent_id]
instance = text_to_disc_instance(tokens, 'real', token_indexers)
instances.append(instance)
# Generate fake batch
num_fake_instances = 0
while num_fake_instances < num_samples:
words, _ = generator.generate()
if not words:
continue
instance = text_to_disc_instance(words, 'fake', token_indexers)
instances.append(instance)
num_fake_instances += 1
return instances
def get_generator_batch(generator: Generator,
token_indexers: Dict[str, TokenIndexer],
batch_size: int) -> List[Tuple[Instance, torch.tensor]]:
instances = []
num_instances = 0
while num_instances < 2 * batch_size:
words, log_likelihood = generator.generate()
if not words:
continue
# HACK: add paddings for the CNN bug
while len(words) <= 5:
words.append(Token(text=DEFAULT_PADDING_TOKEN))
instance = text_to_disc_instance(words, 'fake', token_indexers)
instances.append((instance, log_likelihood))
num_instances += 1
return instances
def get_reward(instance: Instance,
discriminator: Discriminator,
vocab: Vocabulary) -> float:
logits = discriminator.forward_on_instance(instance)['logits']
probs = np.exp(logits) / sum(np.exp(logits)) # softmax
real_label_id = vocab.get_token_index('real', 'labels')
return probs[real_label_id]
def get_reward_chrf(instance: Instance,
train_sentences: List[str],
num_lines=100):
generated = ''.join(token.text for token in instance.fields['tokens'])
line_ids = np.random.choice(len(train_sentences), size=num_lines)
chrf_total = 0.
for line_id in line_ids:
line = train_sentences[line_id]
chrf = sentence_chrf(line, generated, min_len=2, max_len=6, beta=1.,
ignore_whitespace=False)
chrf_total += chrf
return chrf_total / num_lines
def main():
parser = argparse.ArgumentParser(description='SeqGAN training script')
parser.add_argument('--batch-size', type=int, default=256)
parser.add_argument('--g_epochs', type=int, default=10)
parser.add_argument('--embedding_size', type=int, default=32)
parser.add_argument('--hidden_size', type=int, default=256)
parser.add_argument('--num_filters', type=int, default=8)
parser.add_argument('--d_steps', type=int, default=3)
parser.add_argument('--d_epochs', type=int, default=3)
parser.add_argument('--g_lr', type=float, default=1.e-3)
parser.add_argument('--d_lr', type=float, default=1.e-2)
parser.add_argument('--seed', type=int, default=1)
parser.add_argument('--log', type=str, default='log.txt')
args = parser.parse_args()
print(args)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
log_file = open(args.log, mode='w')
all_chars = {END_SYMBOL, START_SYMBOL}
all_chars.update("abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ .,!?'-")
token_counts = {char: 1 for char in all_chars}
label_counts = {'real': 1, 'fake': 1}
vocab = Vocabulary({'tokens': token_counts, 'labels': label_counts})
token_indexers = {'tokens': SingleIdTokenIndexer()}
train_set = read_shakespeare(all_chars=all_chars)
train_sentences = [''.join(token.text for token in tokens) for tokens in train_set]
token_embedding = Embedding(num_embeddings=vocab.get_vocab_size('tokens'),
embedding_dim=args.embedding_size)
embedder = BasicTextFieldEmbedder({"tokens": token_embedding})
generator = Generator(embedder,
embedding_size=args.embedding_size,
hidden_size=args.hidden_size,
max_len=65,
vocab=vocab)
token_embedding = Embedding(num_embeddings=vocab.get_vocab_size('tokens'),
embedding_dim=args.embedding_size)
embedder = BasicTextFieldEmbedder({"tokens": token_embedding})
discriminator = Discriminator(embedder,
embedding_size=args.embedding_size,
num_filters=args.num_filters,
vocab=vocab)
generator_optim = optim.Adam(generator.parameters(), lr=args.g_lr)
discriminator_optim = optim.Adagrad(discriminator.parameters(), lr=args.d_lr)
# pre-train generator
print('Pre-training generator...')
instances = [tokens_to_lm_instance(tokens, token_indexers)
for tokens in train_set]
iterator = BasicIterator(batch_size=args.batch_size)
iterator.index_with(vocab)
trainer = Trainer(model=generator,
optimizer=generator_optim,
iterator=iterator,
train_dataset=instances,
num_epochs=args.g_epochs)
trainer.train()
# pre-train discriminator
print('Pre-training discriminator...')
instances = get_discriminator_batch(
generator, train_set, token_indexers, 10 * args.batch_size)
iterator = BasicIterator(batch_size=args.batch_size)
iterator.index_with(vocab)
trainer = Trainer(model=discriminator,
optimizer=discriminator_optim,
iterator=iterator,
train_dataset=instances,
num_epochs=10)
trainer.train()
for epoch in range(500):
# train generator
generator.zero_grad()
instances = get_generator_batch(
generator, token_indexers, args.batch_size)
log_likelihoods = []
rewards = []
logs = []
for instance, log_likelihood in instances:
reward = get_reward(instance, discriminator, vocab)
reward += get_reward_chrf(instance, train_sentences)
rewards.append(reward)
log_likelihoods.append(log_likelihood)
if len(logs) < 20:
text = ''.join(token.text for token in instance.fields['tokens'])
logs.append(' {:70s} {:4.3f}'.format(text, reward))
baseline = sum(rewards) / len(instances)
avr_loss = sum(-1. * (reward - baseline) * log_likelihood
for reward, log_likelihood in zip(rewards, log_likelihoods))
avr_loss /= len(instances)
avr_loss.backward()
generator_optim.step()
log = 'epoch: {}, loss: {}, avr_reward: {}'.format(epoch, avr_loss, baseline)
print(log)
log_file.write(log + '\n')
for log in logs:
print(log)
log_file.write(log + '\n')
# train discriminator
fake_logs = []
real_logs = []
for d_step in range(args.d_steps):
instances = get_discriminator_batch(
generator, train_set, token_indexers, args.batch_size)
if d_step == 0:
for inst in instances:
text = ''.join(token.text for token in inst.fields['tokens'])
label = inst.fields['label'].label
if label == 'real' and len(real_logs) < 10:
real_logs.append(' {:70s} {}'.format(text, label))
if label == 'fake' and len(fake_logs) < 10:
fake_logs.append(' {:70s} {}'.format(text, label))
iterator = BasicIterator(batch_size=2 * args.batch_size)
iterator.index_with(vocab)
trainer = Trainer(model=discriminator,
optimizer=discriminator_optim,
iterator=iterator,
train_dataset=instances,
num_epochs=args.d_epochs)
trainer.train()
log = 'epoch: {}, step-D'.format(epoch)
print(log)
log_file.write(log + '\n')
for log in real_logs + fake_logs:
print(log)
log_file.write(log + '\n')
log_file.flush()
log_file.close()
if __name__ == '__main__':
main()