-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
196 lines (146 loc) · 7.86 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
import time,os
import torch
import numpy as np
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from transformers import AdamW, get_linear_schedule_with_warmup
from sklearn.model_selection import train_test_split
import argparse
import pandas as pd
import torch.nn as nn
from data.dataset import preprocessing
from model.BertCRF import BertCRF
import random
# from utils.metric import Metrics
from sklearn.metrics import f1_score,accuracy_score,recall_score
import yaml
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")
def seed_anything(seed_value):
np.random.seed(seed_value)
random.seed(seed_value)
os.environ['PYTHONHASHSEED'] = str(seed_value) # 为了禁止hash随机化，使得实验可复现。
torch.manual_seed(seed_value) # 为CPU设置随机种子
torch.cuda.manual_seed(seed_value) # 为当前GPU设置随机种子（只用一块GPU）
torch.cuda.manual_seed_all(seed_value) # 为所有GPU设置随机种子（多块GPU）
torch.backends.cudnn.deterministic = True
def initialize_model(epochs,device,pretrained_bert,types_of_tags,fix_bert,autoCRF,MAX_LEN):
# 定义模型
model = BertCRF(types_of_tags,pretrained_bert,device,fix_bert,autoCRF,MAX_LEN)
model.to(device)
optimizer = AdamW(model.parameters(),
lr=5e-5, # 默认学习率
eps=1e-8 # 默认精度
)
total_steps = len(train_dataloader) * epochs
scheduler = get_linear_schedule_with_warmup(optimizer,
num_warmup_steps=0, # Default value
num_training_steps=total_steps)
return model, optimizer, scheduler
def train(model,train_dataloader,val_dataloader,optimizer,scheduler,epochs,val_gap,device,save_path):
best_f1 = 0.
train_loss_curve = []
for epoch in range(epochs):
model.train()
total_loss = 0
for _,batch in enumerate(train_dataloader):
sentences,masks,tags = tuple(t.to(device) for t in batch)
model.zero_grad()
optimizer.zero_grad()
loss, tag_seq = model.neg_log_likelihood_loss(sentences,masks,tags)
total_loss += loss.item()
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
optimizer.step()
scheduler.step()
train_loss_curve.append(total_loss)
print(f"Epoch {epoch + 1}/{configs['epochs']}, Loss: {total_loss / len(train_dataloader)}")
if epoch % val_gap == 0 or epoch >= epochs:
model.eval()
pred_list = None
label_list = None
mask_list = None
with torch.no_grad():
for _,batch in enumerate(val_dataloader):
sentences,masks,tags = tuple(t.to(device) for t in batch)
tag_seq = model.forward(sentences,masks) #(b,s)
# print(f'Here shape of tag_seq is: {tag_seq.shape}')
if pred_list == None:
pred_list = tag_seq
label_list = tags
mask_list = masks
else:
pred_list = torch.concat((pred_list,tag_seq),0)
label_list = torch.concat((label_list,tags),0)
mask_list = torch.concat((mask_list,masks),0)
# pred_list += tag_seq.cpu().float().tolist()
# label_list += tags.cpu().float().tolist()
pred_list = pred_list.cpu().flatten().tolist()
label_list = label_list.cpu().flatten().tolist()
mask_list = mask_list.cpu().flatten().tolist()
eff_preds = []
eff_labels = []
for i in range(len(mask_list)):
if mask_list[i] == True:
# # 去掉'O'
# if label_list[i] == 16:
# continue
eff_preds.append(pred_list[i])
eff_labels.append(label_list[i])
acc = accuracy_score(eff_labels,eff_preds)
recall = recall_score(eff_labels,eff_preds,average='macro')
f1score = f1_score(eff_labels,eff_preds,average='macro')
print(f'acc is: {acc}')
print(f'recall is: {recall}')
print(f'f1-socre is: {f1score}')
if best_f1 < f1score:
best_f1 = f1score
torch.save(model.state_dict(),os.path.join(save_path,'best.pth'))
torch.save(model.state_dict(),os.path.join(save_path,f'ep{epoch}.pth'))
plt.plot(train_loss_curve)
plt.savefig(os.path.join(save_path,f'loss_curve.png'))
plt.clf()
if __name__ == '__main__':
config_path = './config/train.yaml'
with open(config_path,'r',encoding='utf-8') as f:
configs = yaml.load(f,Loader=yaml.FullLoader)
print(f'EXP Settings: ')
for k,v in configs.items():
print(f'{k}: {v}')
print(f'*'*30)
seed_anything(configs['seed'])
# 创建结果保存路径
if not os.path.exists(os.path.join('./weight',configs['name'])):
os.mkdir(os.path.join('./weight',configs['name']))
if not os.path.exists(os.path.join('./output','train',configs['name'])):
os.mkdir(os.path.join('./output','train',configs['name']))
# 将配置文件保存到模型保存路径
with open(os.path.join('./weight',configs['name'],'train.yaml'),'w') as f:
f.write(yaml.dump(configs,allow_unicode=True))
# 读取训练集和测试集数据
train_sentences, train_masks, train_tag_lists = preprocessing(configs['train_path'],configs['MAX_LEN'],configs['tokenizer_path'],configs['tag2idx_path'])
val_sentences, val_masks, val_tag_lists = preprocessing(configs['val_path'],configs['MAX_LEN'],configs['tokenizer_path'],configs['tag2idx_path'])
# 转化为tensor类型
train_sentences = torch.tensor(train_sentences)
train_masks = torch.tensor(train_masks) > 0.5 # 转为bool
train_tag_lists = torch.tensor(train_tag_lists)
val_sentences = torch.tensor(val_sentences)
val_masks = torch.tensor(val_masks) > 0.5 # 转为bool
val_tag_lists = torch.tensor(val_tag_lists)
# 创建DataLoader
train_dataset = TensorDataset(train_sentences,train_masks,train_tag_lists)
train_sampler = RandomSampler(train_dataset)
train_dataloader = DataLoader(train_dataset,sampler=train_sampler,batch_size=configs['batch_size'],num_workers=configs['num_workers'])
val_dataset = TensorDataset(val_sentences,val_masks,val_tag_lists)
val_sampler = RandomSampler(val_dataset)
val_dataloader = DataLoader(val_dataset,sampler=val_sampler,batch_size=configs['batch_size'],num_workers=configs['num_workers'])
# 加载模型
model, optimizer, scheduler = initialize_model(epochs=configs['epochs'],
device = configs['device'],
pretrained_bert=configs['pretrained_bert'],
types_of_tags = configs['types_of_tags'],
fix_bert = configs['fix_bert'],
autoCRF = configs['autoCRF'],
MAX_LEN = configs['MAX_LEN'])
print("Start training and validating:\n")
train(model,train_dataloader,val_dataloader,optimizer,scheduler,configs['epochs'],configs['val_gap'],configs['device'],os.path.join('./weight',configs['name']))