-
Notifications
You must be signed in to change notification settings - Fork 18
/
sequence_labeling_cws_crf.py
198 lines (162 loc) · 6.55 KB
/
sequence_labeling_cws_crf.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
# -*- coding: utf-8 -*-
# @Date : 2020/8/6
# @Author : mingming.xu
# @Email : xv44586@gmail.com
# @File : sequence_labeling_cws_crf.py
"""
CRF做中文分词（Chinese Word Segment）
数据集 http://sighan.cs.uchicago.edu/bakeoff2005/
官方评测F1：96%
"""
import re
from tqdm import tqdm
import os
import numpy as np
from toolkit4nlp.models import build_transformer_model, Model
from toolkit4nlp.layers import Dense
from toolkit4nlp.utils import ViterbiDecoder
from toolkit4nlp.tokenizers import Tokenizer
from toolkit4nlp.optimizers import Adam
from toolkit4nlp.backend import keras, K
from toolkit4nlp.utils import DataGenerator
from toolkit4nlp.utils import pad_sequences
from toolkit4nlp.layers import ConditionalRandomField
from toolkit4nlp.optimizers import extend_with_gradient_accumulation
train_path = '/home/mingming.xu/datasets/NLP/segment/icwb-2-data/training/pku_training.utf8'
test_path = '/home/mingming.xu/datasets/NLP/segment/icwb-2-data/testing/pku_test.utf8'
data_dir = '/home/mingming.xu/datasets/NLP/segment/icwb-2-data/'
test_result_path = 'test_result.txt'
test_score_path = 'test_score.txt'
bert_config = '/home/mingming.xu/pretrain/NLP/chinese_L-12_H-768_A-12/bert_config.json'
bert_dict = '/home/mingming.xu/pretrain/NLP/chinese_L-12_H-768_A-12/vocab.txt'
bert_checkpoint = '/home/mingming.xu/pretrain/NLP/chinese_L-12_H-768_A-12/bert_model.ckpt'
maxlen = 256
bert_layers = 12
lr_multi = 2 # crf loss 放大
num_labes = 4
lr = 1e-5
batch_size = 16 # 实际梯度在此基础上累加2次，即 batch_size = 32
epochs = 5
tokenizer = Tokenizer(bert_dict, do_lower_case=True)
def load_data(data_path):
items = []
with open(data_path) as f:
for line in f:
chunk = re.split(' +', line.strip())
items.append(chunk)
return items
data = load_data(train_path)
row_nums = list(range(len(data)))
np.random.shuffle(row_nums)
train_data = [data[i] for i in row_nums if i % 10 != 0]
val_data = [data[i] for i in row_nums if i % 10 == 0]
print(len(train_data), len(val_data))
class data_generator(DataGenerator):
def __iter__(self, shuffle=False):
# 0: 单字，1，多字词开头，2，中间，3，末尾
batch_tokens, batch_segs, batch_labels = [], [], []
for is_end, item in self.get_sample(shuffle):
token_ids, labels = [tokenizer._token_start_id], [0]
for word in item:
token_id = tokenizer.encode(word)[0][1:-1]
if len(token_ids) + len(token_id) > maxlen:
break
if len(token_id) == 1:
labels += [0]
else:
labels += [1] + [2] * (len(token_id) - 2) + [3]
token_ids += token_id
token_ids.append(tokenizer._token_end_id)
labels.append(0)
batch_tokens.append(token_ids)
batch_segs.append([0] * len(token_ids))
batch_labels.append(labels)
if len(batch_tokens) >= self.batch_size or is_end:
batch_tokens = pad_sequences(batch_tokens)
batch_segs = pad_sequences(batch_segs)
batch_labels = pad_sequences(batch_labels)
yield [batch_tokens, batch_segs], batch_labels
batch_tokens, batch_segs, batch_labels = [], [], []
model = build_transformer_model(config_path=bert_config, checkpoint_path=bert_checkpoint)
output_layer = 'Transformer-%s-FeedForward-Norm' % (bert_layers - 1)
output = model.get_layer(output_layer).output
output = Dense(num_labes)(output)
CRF = ConditionalRandomField(lr_multi)
output = CRF(output)
model = Model(model.input, output)
model.summary()
class WordSeg(ViterbiDecoder):
def segment(self, data):
tokens = tokenizer.tokenize(data)
while len(tokens) > 512:
tokens.pop(-2)
mapping = tokenizer.rematch(data, tokens)
token_ids = tokenizer.tokens_to_ids(tokens)
segs = [0] * len(token_ids)
pre = model.predict([[token_ids], [segs]])[0]
labels = self.decode(pre)
words = []
for i, label in enumerate(labels[1:-1]):
if label < 2 or len(words)==0:
words.append([i + 1])
else:
words[-1].append(i + 1)
return [data[mapping[w[0]][0]: mapping[w[-1]][-1] + 1] for w in words]
wordseg = WordSeg(trans=K.eval(CRF.trans), starts=[0], ends=[0])
def evaluate(data):
"""简单评测"""
total, right = 1e-10, 1e-10
for true in tqdm(data):
pre = wordseg.segment(''.join(true))
w_pre = set(pre)
w_true = set(true)
total += len(w_true)
right += len(w_pre & w_true)
return right / total
def public_evaluate(test_path, test_result_path, test_score_path):
"""官方评测，结果在score file 的最后几行"""
fw = open(test_result_path, 'w', encoding='utf-8')
with open(test_path, encoding='utf-8') as fr:
for l in tqdm(fr):
l = l.strip()
if l:
l = ' '.join(wordseg.segment(l))
fw.write(l + '\n')
fw.close()
# 运行官方评测脚本
with os.popen(
'{data_dir}/scripts/score {data_dir}/gold/pku_training_words.utf8 {data_dir}/gold/pku_test_gold.utf8 {test_result_path} > {test_score_path}'.format(
data_dir=data_dir, test_result_path=test_result_path, test_score_path=test_score_path)) as p:
p.read()
# 打印结果的最后两行
result = open(test_score_path).readlines()
for line in result[-2:]:
print(line)
class Evaluator(keras.callbacks.Callback):
def __init__(self):
self.best_acc = 0.
def on_epoch_end(self, epoch, logs=None):
trans = K.eval(CRF.trans)
wordseg.trans = trans
print(trans)
acc = evaluate(val_data)
if acc > self.best_acc:
self.best_acc = acc
model.save_weights('./best_model.weights')
print('acc is: {:.3f}, best acc is :{:.4f}'.format(acc, self.best_acc))
def on_train_end(self, logs=None):
model.load_weights('./best_model.weights')
public_evaluate(test_path, test_result_path, test_score_path)
opt = extend_with_gradient_accumulation(Adam)
opt = opt(learning_rate=lr)
model.compile(loss=CRF.sparse_loss, optimizer=opt, metrics=[CRF.sparse_accuracy])
if __name__ == '__main__':
evaluator = Evaluator()
train_genarator = data_generator(train_data, batch_size)
model.fit_generator(
train_genarator.generator(),
steps_per_epoch=len(train_genarator),
epochs=epochs,
callbacks=[evaluator])
else:
model.load_weights('./best_model.weights')