-
Notifications
You must be signed in to change notification settings - Fork 24
/
Copy pathrun.py
223 lines (182 loc) · 9.74 KB
/
run.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
# -*- coding: utf-8 -*-
# @Time : 2018/11/12 15:46
# @Author : Richer
# @File : run.py
from config.config import Config
from tensorflow.contrib.crf import viterbi_decode
from flask import request
from loader import Loader
import tensorflow as tf
from model.IDCNN_model import cnn_model
from model.BiLSTM_model import BiLSTM_model
import numpy as np
import random
from util import Util
from flask import Flask
import os, sys, argparse
import json
class Run(object):
def __init__(self, type='train', model='IDCNN'):
self.config = Config()
self.saver = None
self.util = Util()
self.loader = Loader()
self.model_type = model
self.logger = self.util.get_logger(self.config.log_file)
self.model = cnn_model(self.config) if self.model_type == 'IDCNN' else BiLSTM_model(self.config)
self.ckpt_path = self.config.cnn_ckpt_path if self.model_type == 'IDCNN' else self.config.lstm_ckpt_path
if type == 'train':
self.train()
def save_model(self, sess, epoch):
self.saver = tf.train.Saver()
self.saver.save(sess, self.ckpt_path + '-' + str(epoch))
self.logger.info('save model done')
def _data_preprocess(self, data, zeros, sign=False, char_to_id=[], tag_to_id=[]):
id_to_tag, id_to_char = [], []
train_sentence = self.loader.load_sentences(data, zeros)
print('数据总长度：{}'.format(len(train_sentence)))
self.loader.update_tag_schema(train_sentence, self.config.tag_schema)
if sign:
mappings, char_to_id, id_to_char, tag_to_id, id_to_tag = self.loader.char_mapping(train_sentence,
self.config.lower, sign)
train_data = self.loader.prepare_dataset(train_sentence, char_to_id, tag_to_id, self.config.lower)
else:
train_data = self.loader.prepare_dataset(train_sentence, char_to_id, tag_to_id, self.config.lower)
print('train 预处理后数据长度：{}'.format(len(train_data)))
batch_size = self.config.batch_size if sign else 100
batch_data = self.loader.batch_size_padding(train_data, batch_size)
return batch_data, id_to_tag, tag_to_id, char_to_id
def evaluate(self, sess, trans, data_manager, id_to_tag):
ner_results = self._evaluate(sess, trans, data_manager, id_to_tag)
report = self.util.report_ner(ner_results, self.config.report_file)
return report
def _evaluate(self, sess, trans, data_manager, id_to_tag):
"""
:param sess: session to run the model
:param data: list of data
:param id_to_tag: index to tag name
:return: evaluate result
"""
results = []
trans = trans.eval() # tensor.eval() 相当于 sess.run(self.trans)作用；其实就是执行
for batch in data_manager:
strings, chars, _, tags = batch
lengths, scores = self._run_sess(sess, batch, False)
batch_paths = self._decode(scores, lengths, trans)
for i in range(len(strings)):
result = []
string = strings[i][:lengths[i]]
gold = self.loader.iobes_iob([id_to_tag[int(x)] for x in tags[i][:lengths[i]]])
pred = self.loader.iobes_iob([id_to_tag[int(x)] for x in batch_paths[i][:lengths[i]]])
for char, gold, pred in zip(string, gold, pred):
result.append(" ".join([char, gold, pred]))
results.append(result)
return results
def _decode(self, logits, lengths, matrix):
"""
:param logits: [batch_size, num_steps, num_tags]float32, logits
:param lengths: [batch_size]int32, real length of each sequence
:param matrix: transaction matrix for inference
:return:
"""
paths = []
small = -1000.0
start = np.asarray([[small] * self.config.num_tags + [0]]) # 初始化一个
for score, length in zip(logits, lengths):
score = score[:length]
pad = small * np.ones([length, 1]) # 创建一个字符长度是 输入字长度维度元素为1的np数组
logits = np.concatenate([score, pad], axis=1)
logits = np.concatenate([start, logits], axis=0)
path, _ = viterbi_decode(logits, matrix)
paths.append(path[1:])
return paths
def _run_sess(self, sess, batch, is_train):
self._create_feed_dict(batch)
if is_train:
loss, train_op, lengths, trans, global_step, learn_rate = sess.run(
[self.model.loss, self.model.train_op, self.model.lengths, self.model.trans, self.model.global_step,
self.model.lr], self.feed_dict)
return loss, lengths, trans, global_step, learn_rate
else:
lengths, logits = sess.run([self.model.lengths, self.model.logits], self.feed_dict)
return lengths, logits
def _create_feed_dict(self, batch, is_train=True):
_, chars, segs, tags = batch
self.feed_dict = {
self.model.char_inputs: np.asarray(chars),
self.model.seg_inputs: np.asarray(segs),
self.model.dropout: 1.0,
}
if is_train:
self.feed_dict[self.model.targets] = np.asarray(tags)
self.feed_dict[self.model.dropout] = self.config.dropout
def _evaluate_line(self, sess, inputs, id_to_tag):
'''
:param sess:
:param inputs:
:param id_to_tag:
:return:
'''
trans = self.model.trans.eval(session=sess)
lengths, scores = self._run_sess(sess, inputs, False)
batch_paths = self._decode(scores, lengths, trans)
tags = [id_to_tag[idx] for idx in batch_paths[0]]
return self.util.result_to_json(inputs[0][0], tags)
def train(self):
batch_data, id_to_tag, tag_to_id, char_to_id = self._data_preprocess(self.config.train_file, self.config.zeros,
True)
self.logger.info('train data prepare done')
dev_batch_data, _, _, _ = self._data_preprocess(self.config.dev_file, self.config.zeros, False, char_to_id,
tag_to_id)
self.logger.info('dev data prepare done')
self.logger.info('start train......')
batch_len = len(batch_data)
tf_config = tf.ConfigProto()
# tf_config.gpu_options.allow_growth = True 这个是动态允许使用gpu空间
tf_config.gpu_options.per_process_gpu_memory_fraction = 0.8
with tf.Session(config=tf_config) as sess:
sess.run(tf.global_variables_initializer())
for epoch in range(self.config.max_epoch):
lr = ''
ls = ''
random.shuffle(batch_data)
for step in range(batch_len):
loss, lengths, trans, global_step, learn_rate = self._run_sess(sess, batch_data[step], True)
if step == (batch_len - 1):
lr, ls = learn_rate, loss
if (int(step) + 1) % self.config.steps_check == 0:
self.logger.info(
' epoch:{}, step/total_batch:{}/{}, global_step:{}, learn_rate:{}, loss:{}'.format(epoch,
step,
batch_len,
global_step,
learn_rate,
loss))
if (epoch + 1) % 2 == 0:
print('*' * 50)
report = self.evaluate(sess, self.model.trans, dev_batch_data, id_to_tag)
self.logger.info(report[1].strip())
self.logger.info('dev: epoch:{}, learn_rate:{}, loss:{}'.format(epoch, lr, ls))
if (int(epoch) + 1) % 20 == 0:
self.save_model(sess, epoch)
def online(self, inputs):
if not inputs:
return json.dumps({'result':'error'})
with open('./data/id_to_tag.txt', 'r', encoding='utf-8') as tag, open('./data/char_to_id.txt', 'r',
encoding='utf-8') as char:
id_to_tag = {int(line.strip().split(':')[0]): line.strip().split(":")[1] for line in tag.readlines()}
char_to_id = {s[0:s.rfind(':')].strip(): int(s[s.rfind(':') + 1:].strip()) for s in char.readlines()}
self.saver = tf.train.Saver()
tf_config = tf.ConfigProto() # 实例化一个设置GPU的对象 函数用在创建session的时候，用来对session进行参数配置
tf_config.gpu_options.per_process_gpu_memory_fraction = 0.8
with tf.Session(config=tf_config) as sess:
ckpt = tf.train.get_checkpoint_state(self.ckpt_path)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
self.logger.info('restore model')
self.saver.restore(sess, ckpt.model_checkpoint_path)
data = self.loader.input_from_line(inputs, char_to_id)
result = self._evaluate_line(sess, data, id_to_tag)
return json.dumps(result)
#如果是训练模型开启以下注释项执行项; 其中模型可选择IDCNN或者BILSTM
# if __name__ == '__main__':
# main = Run('train', 'IDCNN')