-
Notifications
You must be signed in to change notification settings - Fork 70
/
train_translator.py
172 lines (139 loc) · 7.36 KB
/
train_translator.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
import tensorflow as tf
import numpy as np
import argparse
import model_config
import data_loader
from ByteNet import translator
import utils
import shutil
import time
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--learning_rate', type=float, default=0.001,
help='Learning Rate')
parser.add_argument('--batch_size', type=int, default=8,
help='Learning Rate')
parser.add_argument('--bucket_quant', type=int, default=50,
help='Learning Rate')
parser.add_argument('--max_epochs', type=int, default=1000,
help='Max Epochs')
parser.add_argument('--beta1', type=float, default=0.5,
help='Momentum for Adam Update')
parser.add_argument('--resume_model', type=str, default=None,
help='Pre-Trained Model Path, to resume from')
parser.add_argument('--source_file', type=str, default='Data/MachineTranslation/news-commentary-v11.de-en.de',
help='Source File')
parser.add_argument('--target_file', type=str, default='Data/MachineTranslation/news-commentary-v11.de-en.en',
help='Target File')
parser.add_argument('--sample_every', type=int, default=500,
help='Sample generator output evry x steps')
parser.add_argument('--summary_every', type=int, default=50,
help='Sample generator output evry x steps')
parser.add_argument('--top_k', type=int, default=5,
help='Sample from top k predictions')
parser.add_argument('--resume_from_bucket', type=int, default=0,
help='Resume From Bucket')
args = parser.parse_args()
data_loader_options = {
'model_type' : 'translation',
'source_file' : args.source_file,
'target_file' : args.target_file,
'bucket_quant' : args.bucket_quant,
}
dl = data_loader.Data_Loader(data_loader_options)
buckets, source_vocab, target_vocab = dl.load_translation_data()
print "Number Of Buckets", len(buckets)
config = model_config.translator_config
model_options = {
'source_vocab_size' : len(source_vocab),
'target_vocab_size' : len(target_vocab),
'residual_channels' : config['residual_channels'],
'decoder_dilations' : config['decoder_dilations'],
'encoder_dilations' : config['encoder_dilations'],
'decoder_filter_width' : config['decoder_filter_width'],
'encoder_filter_width' : config['encoder_filter_width'],
}
translator_model = translator.ByteNet_Translator( model_options )
translator_model.build_model()
optim = tf.train.AdamOptimizer(
args.learning_rate,
beta1 = args.beta1).minimize(translator_model.loss)
translator_model.build_translator(reuse = True)
merged_summary = tf.summary.merge_all()
sess = tf.InteractiveSession()
tf.initialize_all_variables().run()
saver = tf.train.Saver()
if args.resume_model:
saver.restore(sess, args.resume_model)
shutil.rmtree('Data/tb_summaries/translator_model')
train_writer = tf.summary.FileWriter('Data/tb_summaries/translator_model', sess.graph)
bucket_sizes = [bucket_size for bucket_size in buckets]
bucket_sizes.sort()
step = 0
batch_size = args.batch_size
for epoch in range(args.max_epochs):
for bucket_size in bucket_sizes:
if epoch == 0 and bucket_size < args.resume_from_bucket:
continue
batch_no = 0
while (batch_no + 1) * batch_size < len(buckets[bucket_size]):
start = time.clock()
source, target = dl.get_batch_from_pairs(
buckets[bucket_size][batch_no * batch_size : (batch_no+1) * batch_size]
)
_, loss, prediction = sess.run(
[optim, translator_model.loss, translator_model.arg_max_prediction],
feed_dict = {
translator_model.source_sentence : source,
translator_model.target_sentence : target,
})
end = time.clock()
print "LOSS: {}\tEPOCH: {}\tBATCH_NO: {}\t STEP:{}\t total_batches:{}\t bucket_size:{}".format(
loss, epoch, batch_no, step, len(buckets[bucket_size])/args.batch_size, bucket_size)
print "TIME FOR BATCH", end - start
print "TIME FOR BUCKET (mins)", (end - start) * (len(buckets[bucket_size])/args.batch_size)/60.0
batch_no += 1
step += 1
if step % args.summary_every == 0:
[summary] = sess.run([merged_summary], feed_dict = {
translator_model.source_sentence : source,
translator_model.target_sentence : target,
})
train_writer.add_summary(summary, step)
print "******"
print "Source ", dl.inidices_to_string(source[0], source_vocab)
print "---------"
print "Target ", dl.inidices_to_string(target[0], target_vocab)
print "----------"
print "Prediction ",dl.inidices_to_string(prediction[0:bucket_size], target_vocab)
print "******"
if step % args.sample_every == 0:
log_file = open('Data/translator_sample.txt', 'wb')
generated_target = target[:,0:1]
for col in range(bucket_size):
[probs] = sess.run([translator_model.t_probs],
feed_dict = {
translator_model.t_source_sentence : source,
translator_model.t_target_sentence : generated_target,
})
curr_preds = []
for bi in range(probs.shape[0]):
pred_word = utils.sample_top(probs[bi][-1], top_k = args.top_k )
curr_preds.append(pred_word)
generated_target = np.insert(generated_target, generated_target.shape[1], curr_preds, axis = 1)
for bi in range(probs.shape[0]):
print col, dl.inidices_to_string(generated_target[bi], target_vocab)
print col, dl.inidices_to_string(target[bi], target_vocab)
print "***************"
if col == bucket_size - 1:
try:
log_file.write("Predicted: " + dl.inidices_to_string(generated_target[bi], target_vocab) + '\n')
log_file.write("Actual Target: " + dl.inidices_to_string(target[bi], target_vocab) + '\n')
log_file.write("Actual Source: " + dl.inidices_to_string(source[bi], source_vocab) + '\n *******')
except:
pass
print "***************"
log_file.close()
save_path = saver.save(sess, "Data/Models/translation_model/model_epoch_{}_{}.ckpt".format(epoch, bucket_size))
if __name__ == '__main__':
main()