/
train.py
executable file
·217 lines (195 loc) · 8.62 KB
/
train.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
import argparse
import configparser
import os
import glob
import logging
from logging import getLogger
import numpy as np
np.set_printoptions(precision=3)
# os.environ["CHAINER_TYPE_CHECK"] = "0"
import chainer
import dataset
import converter
import iterator
from evaluate import Evaluate
from hi_seq2seq import HiSeq2SeqModel
from word_encoder import WordEnc
from word_decoder import WordDec
from sent_encoder import SentEnc
from sent_decoder import SentDec
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('model_dir')
parser.add_argument('--batch', '-b', type=int, default=32)
parser.add_argument('--epoch', '-e', type=int, default=20)
parser.add_argument('--interval', '-i', type=int, default=10000)
parser.add_argument('--gpu', '-g', type=int, default=-1)
args = parser.parse_args()
return args
def main():
args = parse_args()
model_dir = args.model_dir
"""LOAD CONFIG FILE"""
config_files = glob.glob(os.path.join(model_dir, '*.ini'))
assert len(config_files) == 1, 'Put only one config file in the directory'
config_file = config_files[0]
config = configparser.ConfigParser()
config.read(config_file)
"""LOGGER"""
logger = getLogger(__name__)
logger.setLevel(logging.INFO)
formatter = logging.Formatter('[%(asctime)s] %(message)s')
sh = logging.StreamHandler()
sh.setLevel(logging.INFO)
sh.setFormatter(formatter)
logger.addHandler(sh)
log_file = model_dir + 'log.txt'
fh = logging.FileHandler(log_file)
fh.setLevel(logging.INFO)
fh.setFormatter(formatter)
logger.addHandler(fh)
logger.info('[Training start] logging to {}'.format(log_file))
"""PARAMATER"""
embed_size = int(config['Parameter']['embed_size'])
hidden_size = int(config['Parameter']['hidden_size'])
dropout_ratio = float(config['Parameter']['dropout'])
weight_decay = float(config['Parameter']['weight_decay'])
gradclip = float(config['Parameter']['gradclip'])
vocab_type = config['Parameter']['vocab_type']
vocab_size = int(config['Parameter']['vocab_size'])
"""TRINING DETAIL"""
gpu_id = args.gpu
n_epoch = args.epoch
batch_size = args.batch
interval = args.interval
"""DATASET"""
train_src_file = config['Dataset']['train_src_file']
train_trg_file = config['Dataset']['train_trg_file']
valid_src_file = config['Dataset']['valid_src_file']
valid_trg_file = config['Dataset']['valid_trg_file']
test_src_file = config['Dataset']['test_src_file']
correct_txt_file = config['Dataset']['correct_txt_file']
train_data_size = dataset.data_size(train_trg_file)
valid_data_size = dataset.data_size(valid_trg_file)
logger.info('train size: {0}, valid size: {1}'.format(train_data_size, valid_data_size))
if vocab_type == 'normal':
init_vocab = {'<unk>': 0, '<s>': 1, '</s>': 2, '<eod>': 3}
vocab = dataset.VocabNormal()
vocab.make_vocab(train_src_file, train_trg_file, init_vocab, vocab_size, freq=0)
dataset.save_pickle(model_dir + 'src_vocab.pkl', vocab.src_vocab)
dataset.save_pickle(model_dir + 'trg_vocab.pkl', vocab.trg_vocab)
sos = vocab.src_vocab['<s>']
eos = vocab.src_vocab['</s>']
eod = vocab.src_vocab['<eod>']
elif vocab_type == 'subword':
vocab = dataset.VocabSubword()
if os.path.isfile(model_dir + 'src_vocab.sub.model') and os.path.isfile(model_dir + 'trg_vocab.sub.model'):
vocab.load_vocab(model_dir + 'src_vocab.sub.model', model_dir + 'trg_vocab.sub.model')
else:
vocab.make_vocab(train_trg_file + '.sub', train_trg_file + '.sub', model_dir, vocab_size)
sos = vocab.src_vocab.PieceToId('<s>')
eos = vocab.src_vocab.PieceToId('</s>')
eod = vocab.src_vocab.PieceToId('<eod>')
src_vocab_size = len(vocab.src_vocab)
trg_vocab_size = len(vocab.trg_vocab)
logger.info('src_vocab size: {}, trg_vocab size: {}'.format(src_vocab_size, trg_vocab_size))
train_iter = iterator.Iterator(train_src_file, train_trg_file, batch_size, sort=True, shuffle=True)
# train_iter = iterator.Iterator(train_src_file, train_trg_file, batch_size, sort=False, shuffle=False)
valid_iter = iterator.Iterator(valid_src_file, valid_trg_file, batch_size, sort=False, shuffle=False)
evaluater = Evaluate(correct_txt_file)
test_iter = iterator.Iterator(test_src_file, test_src_file, batch_size, sort=False, shuffle=False)
"""MODEL"""
model = HiSeq2SeqModel(
WordEnc(src_vocab_size, embed_size, hidden_size, dropout_ratio),
WordDec(trg_vocab_size, embed_size, hidden_size, dropout_ratio),
SentEnc(hidden_size, dropout_ratio),
SentDec(hidden_size, dropout_ratio),
sos, eos, eod)
"""OPTIMIZER"""
optimizer = chainer.optimizers.Adam()
optimizer.setup(model)
optimizer.add_hook(chainer.optimizer.GradientClipping(gradclip))
optimizer.add_hook(chainer.optimizer.WeightDecay(weight_decay))
"""GPU"""
if gpu_id >= 0:
logger.info('Use GPU')
chainer.cuda.get_device_from_id(gpu_id).use()
model.to_gpu()
"""TRAIN"""
sum_loss = 0
loss_dic = {}
for epoch in range(1, n_epoch + 1):
for i, batch in enumerate(train_iter.generate(), start=1):
print(batch)
exit()
batch = vocab.convert2label(batch)
data = converter.convert(batch, gpu_id)
loss = optimizer.target(*data)
sum_loss += loss.data
optimizer.target.cleargrads()
loss.backward()
optimizer.update()
if i % interval == 0:
logger.info('E{} ## iteration:{}, loss:{}'.format(epoch, i, sum_loss))
sum_loss = 0
chainer.serializers.save_npz(model_dir + 'model_epoch_{}.npz'.format(epoch), model)
# chainer.serializers.save_npz(model_dir + 'optimizer_epoch{0}.npz'.format(epoch), optimizer)
"""EVALUATE"""
valid_loss = 0
for batch in valid_iter.generate():
batch = vocab.convert2label(batch)
data = converter.convert(batch, gpu_id)
with chainer.no_backprop_mode(), chainer.using_config('train', False):
valid_loss += optimizer.target(*data).data
logger.info('E{} ## val loss:{}'.format(epoch, valid_loss))
loss_dic[epoch] = valid_loss
"""TEST"""
output = []
for batch in test_iter.generate():
# batch: (articlesのリスト, abstracts_sosのリスト, abstracts_eosのリスト)タプル
batch = vocab.convert2label(batch)
data = converter.convert(batch, gpu_id)
"""
out: [(sent, attn), (sent, attn), ...] <-バッチサイズ
sent: decodeされた文のリスト
attn: 各文のdecode時のattentionのリスト
"""
with chainer.no_backprop_mode(), chainer.using_config('train', False):
out = model.generate(data[0], data[3])
output.extend(out)
res_decode = []
res_attn = []
for o in output:
sent, attn = o
sentence = dataset.to_list(sent)
sentence = dataset.eod_truncate(sentence, eod)
sent_num = len(sentence)
sentence = [dataset.eos_truncate(s, eos) for s in sentence]
sentence = [vocab.label2word(s) for s in sentence]
sentence = dataset.join_sentences(sentence)
res_decode.append(sentence)
attn = np.sum(np.array(attn[:sent_num]), axis=0) / sent_num
res_attn.append(attn)
rank_list = evaluater.rank(res_attn)
single = evaluater.single(rank_list)
multiple = evaluater.multiple(rank_list)
logger.info('E{} ## precision'.format(epoch))
logger.info('single: {} | {}'.format(single[0], single[1]))
logger.info('multi : {} | {}'.format(multiple[0], multiple[1]))
with open(model_dir + 'model_epoch_{}.hypo'.format(epoch), 'w')as f:
[f.write(r + '\n') for r in res_decode]
with open(model_dir + 'model_epoch_{}.attn'.format(epoch), 'w')as f:
[f.write('{}\n'.format(r)) for r in res_attn]
with open(model_dir + 'model_epoch_{}.prec'.format(epoch), 'w')as f:
f.write('single\n')
f.write(single[0] + '\n')
f.write(single[1] + '\n')
f.write('multiple\n')
f.write(multiple[0] + '\n')
f.write(multiple[1] + '\n')
"""MODEL SAVE"""
best_epoch = min(loss_dic, key=(lambda x: loss_dic[x]))
logger.info('best_epoch:{0}'.format(best_epoch))
chainer.serializers.save_npz(model_dir + 'best_model.npz', model)
if __name__ == '__main__':
main()