-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtask_nmt_iwslt17_nl2en.py
180 lines (151 loc) · 5.72 KB
/
task_nmt_iwslt17_nl2en.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
#! -*- coding: utf-8 -*-
# take bert for NMT task and employ the UNILM seq2seq method
# refer to bert4keras：https://github.com/bojone/bert4keras
from __future__ import print_function
import os
os.environ['TF_KERAS']= '1'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import numpy as np
from tqdm import tqdm
from bert4keras.backend import keras, K
from bert4keras.layers import Loss
from bert4keras.models import build_transformer_model
from bert4keras.tokenizers import Tokenizer, load_vocab
from bert4keras.optimizers import Adam
from bert4keras.snippets import sequence_padding, open
from bert4keras.snippets import DataGenerator, AutoRegressiveDecoder
from keras.models import Model
from rouge import Rouge # pip install rouge
from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction
import tensorflow as tf
# hpyer-parameters
maxlen = 200
batch_size = 32
epochs = 8
# bert config
#multi-language BERT pretrained model
config_path = '/models/multi_cased_L-12_H-768_A-12/bert_config.json'
checkpoint_path = '/models/multi_cased_L-12_H-768_A-12/bert_model.ckpt'
dict_path = '/models/multi_cased_L-12_H-768_A-12/vocab.txt'
def load_data(filename):
"""load data
item：(nl, en)
"""
D = []
with open(filename, encoding='utf-8') as f:
for l in f:
title, content = l.strip().split('\t')
D.append((title, content))
return D
# load datasets
train_data = load_data('datasets/iwslt2017/ennl/corpus_nlen.tsv')
valid_data = load_data('datasets/iwslt2017/ennl/dev2010_nlen.tsv')
#test_data = load_data('datasets/iwslt2016/test_lowcased.tsv')
# load dict and tokenize
token_dict, keep_tokens = load_vocab(
dict_path=dict_path,
simplified=True,
startswith=['[PAD]', '[UNK]', '[CLS]', '[SEP]'],
)
# token_dict = load_vocab(dict_path)
tokenizer = Tokenizer(token_dict, do_lower_case=True)
class data_generator(DataGenerator):
"""data generator
"""
def __iter__(self, random=False):
batch_token_ids, batch_segment_ids = [], []
for is_end, (title, content) in self.sample(random):
token_ids, segment_ids = tokenizer.encode(
title, content, maxlen=maxlen
)
batch_token_ids.append(token_ids)
batch_segment_ids.append(segment_ids)
if len(batch_token_ids) == self.batch_size or is_end:
batch_token_ids = sequence_padding(batch_token_ids)
batch_segment_ids = sequence_padding(batch_segment_ids)
yield ([batch_token_ids, batch_segment_ids],None )
batch_token_ids, batch_segment_ids = [], []
class CrossEntropy(Loss):
"""cross-entropy as loss
"""
def compute_loss(self, inputs, mask=None):
y_true, y_mask, y_pred = inputs
y_true = y_true[:, 1:] # # target token_ids
y_mask = y_mask[:, 1:] # segment_ids
y_pred = y_pred[:, :-1] # predicted sequence
loss = K.sparse_categorical_crossentropy(y_true, y_pred)
loss = K.sum(loss * y_mask) / K.sum(y_mask)
return loss
model = build_transformer_model(
config_path,
checkpoint_path,
application='unilm',
# keep_tokens=keep_tokens
)
output = CrossEntropy(2)(model.inputs + model.outputs)
model = Model(model.inputs, output)
model.compile(optimizer=Adam(1e-5))
model.summary()
class AutoTitle(AutoRegressiveDecoder):
"""seq2seq decoder
"""
@AutoRegressiveDecoder.wraps(default_rtype='probas')
def predict(self, inputs, output_ids, states):
token_ids, segment_ids = inputs
token_ids = np.concatenate([token_ids, output_ids], 1)
segment_ids = np.concatenate([segment_ids, np.ones_like(output_ids)], 1)
return self.last_token(model).predict([token_ids, segment_ids])
def generate(self, text, topk=1):
max_c_len = maxlen - self.maxlen
token_ids, segment_ids = tokenizer.encode(text, maxlen=max_c_len)
output_ids = self.beam_search([token_ids, segment_ids],
topk=topk) # 基于beam search
return tokenizer.decode(output_ids)
autotitle = AutoTitle(start_id=None, end_id=tokenizer._token_end_id, maxlen=64) #32,64,80 64 best
class Evaluator(keras.callbacks.Callback):
"""evaluate and save the model
"""
def __init__(self):
self.lowest = 1e10
def on_epoch_end(self, epoch, logs=None):
# save the best model
if logs['loss'] <= self.lowest:
self.lowest = logs['loss']
model.save_weights('./iwslt2017_nl2en_model/best_model_nlen.weights')
#train
if __name__ == '__main__':
evaluator = Evaluator()
train_generator = data_generator(train_data, batch_size)
# d = train_generator.forfit()
# print(d.__next__())
# print(d.__next__())
model.fit(
train_generator.forfit(),
steps_per_epoch=len(train_generator),
epochs=epochs,
callbacks=[evaluator]
)
#
else:
model.load_weights('./iwslt2017_nl2en_model/best_model_nlen.weights')
# '''''''''''''''''''''test'''''''''''''''''
model.load_weights('./iwslt2017_nl2en_model/best_model_nlen.weights')
enss = []
for (de,en) in valid_data:
ens = autotitle.generate(de,topk=4)
enss.append(ens)
import codecs
def write_trans_result(filename):
with codecs.open(filename,'w') as f:
f.write('\n'.join(enss))
write_trans_result('datasets/iwslt2017/ennl/dev2010.en.trans')
#---------------------get source file-------------
import codecs#def get_src(filename1,filename2):
D = []
with open(filename1) as f:
for l in f:
title, content = l.strip().split('\t')
D.append(content)
with open(filename2,'w') as f1:
f1.write('\n'.join(D))
get_src('datasets/iwslt2017/ennl/dev2010_nlen.tsv','datasets/iwslt2017/ennl/dev2010.src.en')