-
Notifications
You must be signed in to change notification settings - Fork 13
/
baseline.py
190 lines (162 loc) · 5.83 KB
/
baseline.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
#! -*- coding: utf-8 -*-
# 2021搜狐校园文本匹配算法大赛baseline
# 直接用 RoFormer + Cond LayerNorm 融合为一个模型
import json
import numpy as np
from bert4keras.backend import keras
from bert4keras.models import build_transformer_model
from bert4keras.tokenizers import Tokenizer
from bert4keras.optimizers import Adam, extend_with_exponential_moving_average
from bert4keras.snippets import sequence_padding, DataGenerator
from bert4keras.snippets import open, to_array
from keras.layers import Input, Embedding, Reshape, GlobalAveragePooling1D, Dense
from keras.models import Model
from tqdm import tqdm
import jieba
jieba.initialize()
# 基本信息
maxlen = 512
epochs = 5
batch_size = 16
learing_rate = 2e-5
# bert配置
config_path = '/root/kg/bert/chinese_roformer_L-12_H-768_A-12/bert_config.json'
checkpoint_path = '/root/kg/bert/chinese_roformer_L-12_H-768_A-12/bert_model.ckpt'
dict_path = '/root/kg/bert/chinese_roformer_L-12_H-768_A-12/vocab.txt'
variants = [
u'短短匹配A类',
u'短短匹配B类',
u'短长匹配A类',
u'短长匹配B类',
u'长长匹配A类',
u'长长匹配B类',
]
# 读取数据
train_data, valid_data, test_data = [], [], []
for i, var in enumerate(variants):
key = 'labelA' if 'A' in var else 'labelB'
fs = [
'../datasets/sohu2021_open_data_clean/%s/train.txt' % var,
'../datasets/round2/%s.txt' % var
]
for f in fs:
with open(f) as f:
for l in f:
l = json.loads(l)
train_data.append((i, l['source'], l['target'], int(l[key])))
f = '../datasets/sohu2021_open_data_clean/%s/valid.txt' % var
with open(f) as f:
for l in f:
l = json.loads(l)
valid_data.append((i, l['source'], l['target'], int(l[key])))
f = '../datasets/sohu2021_open_data_clean/%s/test_with_id.txt' % var
with open(f) as f:
for l in f:
l = json.loads(l)
test_data.append((i, l['source'], l['target'], l['id']))
# 建立分词器
tokenizer = Tokenizer(
dict_path,
do_lower_case=True,
pre_tokenize=lambda s: jieba.lcut(s, HMM=False)
)
class data_generator(DataGenerator):
"""数据生成器
"""
def __iter__(self, random=False):
batch_token_ids, batch_segment_ids = [], []
batch_conds, batch_labels = [], []
for is_end, (cond, source, target, label) in self.sample(random):
token_ids, segment_ids = tokenizer.encode(
source, target, maxlen=maxlen
)
batch_token_ids.append(token_ids)
batch_segment_ids.append(segment_ids)
batch_conds.append([cond])
batch_labels.append([label])
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)
batch_conds = sequence_padding(batch_conds)
batch_labels = sequence_padding(batch_labels)
yield [
batch_token_ids, batch_segment_ids, batch_conds
], batch_labels
batch_token_ids, batch_segment_ids = [], []
batch_conds, batch_labels = [], []
c_in = Input(shape=(1,))
c = Embedding(len(variants), 128)(c_in)
c = Reshape((128,))(c)
model = build_transformer_model(
config_path,
checkpoint_path,
model='roformer',
layer_norm_cond=c,
additional_input_layers=c_in
)
output = GlobalAveragePooling1D()(model.output)
output = Dense(2, activation='softmax')(output)
model = Model(model.inputs, output)
model.summary()
AdamEMA = extend_with_exponential_moving_average(Adam, name='AdamEMA')
optimizer = AdamEMA(learing_rate, ema_momentum=0.9999)
model.compile(
loss='sparse_categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy']
)
# 转换数据集
train_generator = data_generator(train_data, batch_size)
valid_generator = data_generator(valid_data, batch_size)
test_generator = data_generator(test_data, batch_size)
def evaluate(data):
"""评测函数(A、B两类分别算F1然后求平均)
"""
total_a, right_a = 0., 0.
total_b, right_b = 0., 0.
for x_true, y_true in tqdm(data):
y_pred = model.predict(x_true).argmax(axis=1)
y_true = y_true[:, 0]
flag = x_true[2][:, 0] % 2
total_a += ((y_pred + y_true) * (flag == 0)).sum()
right_a += ((y_pred * y_true) * (flag == 0)).sum()
total_b += ((y_pred + y_true) * (flag == 1)).sum()
right_b += ((y_pred * y_true) * (flag == 1)).sum()
f1_a = 2.0 * right_a / total_a
f1_b = 2.0 * right_b / total_b
return {'f1': (f1_a + f1_b) / 2, 'f1_a': f1_a, 'f1_b': f1_b}
def predict_test(filename):
"""测试集预测到文件
"""
with open(filename, 'w') as f:
f.write('id,label\n')
for x_true, y_true in tqdm(test_generator):
y_pred = model.predict(x_true).argmax(axis=1)
y_true = y_true[:, 0]
for id, y in zip(y_true, y_pred):
f.write('%s,%s\n' % (id, y))
class Evaluator(keras.callbacks.Callback):
"""评估与保存
"""
def __init__(self):
self.best_val_f1 = 0.
def on_epoch_end(self, epoch, logs=None):
optimizer.apply_ema_weights()
metrics = evaluate(valid_generator)
if metrics['f1'] > self.best_val_f1:
self.best_val_f1 = metrics['f1']
model.save_weights('best_model.weights')
optimizer.reset_old_weights()
metrics['best_f1'] = self.best_val_f1
print(metrics)
if __name__ == '__main__':
evaluator = Evaluator()
model.fit(
train_generator.forfit(),
steps_per_epoch=len(train_generator),
epochs=epochs,
callbacks=[evaluator]
)
else:
model.load_weights('best_model.weights')
# predict_test('test.csv')