-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsksmt_train.py
274 lines (244 loc) · 12.9 KB
/
sksmt_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
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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
# -*- coding:utf-8 -*-
# Description TODO
# author Rover Email:[email protected]
# version 1.0
# date 2021/11/17 11:00
import argparse
import os
import logging
import torch
from torch import nn
from Models import SKSMT,AFFGCN #模型
from Utils import SKSMTDatesetReader,SKSMTBatchIteraotr
from torchsummary import summary #输出模型详细信息
import math
from sklearn import metrics
logger = logging.getLogger(__name__)
class Instructor:
def __init__(self,opt):
self.opt=opt #配置
dataset=SKSMTDatesetReader(dataset=opt.dataset, embed_dim=opt.embed_dim, max_seq_len = opt.max_seq_len)
self.train_data_loader = SKSMTBatchIteraotr(data=dataset.semeval.train_data, batch_size=opt.batch_size, shuffle=True)
self.test_data_loader = SKSMTBatchIteraotr(data=dataset.semeval.test_data, batch_size=opt.batch_size, shuffle=False)
self.model = opt.model_class(opt,dataset.semeval.vocab_size).to(opt.device)
# summary(self.model,(16,128,300))
self._print_args()
self.global_f1 = 0. #全局的f1
if torch.cuda.is_available():
print('cuda memory allocated:', torch.cuda.memory_allocated(device=opt.device.index))
def _print_args(self):
'''
输出一些训练参数的信息
:return:
'''
n_trainable_params, n_nontrainable_params = 0, 0
for p in self.model.parameters():
n_params = torch.prod(torch.tensor(p.shape)).item()
if p.requires_grad:
n_trainable_params += n_params #需要训练的参数
else:
n_nontrainable_params += n_params #不参与训练的参数
print('n_trainable_params: {0}, n_nontrainable_params: {1}'.format(n_trainable_params, n_nontrainable_params))
print('> training arguments:')
for arg in vars(self.opt):
print('>>> {0}: {1}'.format(arg, getattr(self.opt, arg)))
def run(self,repeats=3):
'''
:param repeats:
:return:
'''
loss_function=nn.CrossEntropyLoss() #损失函数
_params=filter(lambda p: p.requires_grad, self.model.parameters()) #将需要训练的参数过滤出来
optimizer = self.opt.optimizer(_params, lr=self.opt.learning_rate, weight_decay=self.opt.l2reg)#优化器
if not os.path.exists('log/'):
os.mkdir('log/')
f_out = open('log/'+self.opt.model_name+'_'+self.opt.dataset+'_val.txt', 'a+', encoding='utf-8')
f_out.write("\n"+"#"*100)
test_acc=[]
test_f1=[]
test_pre = []
test_recall = []
max_test_acc_avg = 0
max_test_f1_avg = 0
for i in range(repeats): #重复训练repeats次。
print('repeat: ', (i+1))
f_out.write('\nrepeat: '+str(i+1))
self._reset_params() #给参数初始化
max_test_acc, max_test_f1 ,max_test_precision , max_test_recall= self._train(loss_function, optimizer) #训练
test_acc.append(max_test_acc)
test_f1.append(max_test_f1)
test_pre.append(max_test_precision)
test_recall.append(max_test_recall)
print('max_test_acc: {0} max_test_f1: {1}'.format(max_test_acc, max_test_f1))
f_out.write('\n\tmax_test_acc: {0}, max_test_f1: {1}'.format(max_test_acc, max_test_f1))
max_test_acc_avg += max_test_acc
max_test_f1_avg += max_test_f1
print('#' * 100)
print("max_test_acc_avg:", max_test_acc_avg/repeats)
print("max_test_f1_avg:", max_test_f1_avg/repeats)
f_out.write('\n\tmax_test_acc_avg: {0}, max_test_f1_avg: {1}'.format(max_test_acc_avg/repeats, max_test_f1_avg/repeats))
for i,j,k,p in zip(test_acc,test_f1,test_pre,test_recall):
print('max_test_acc: {0} max_test_f1: {1} max_test_pre: {2} max_test_recall: {3}'.format(i, j, k, p))
print(self.opt.log_info)
f_out.close()
def _train(self,loss_funciton,optimizer):
max_test_acc = 0
max_test_f1 = 0
max_test_precision = 0
max_test_recall = 0
global_step = 0
continue_not_increase = 0 #判断早停的依据
for epoch in range(self.opt.num_epoch): #epoch
print('>' * 100)
print('epoch: ', epoch)
n_correct, n_total = 0, 0
increase_flag = False
for i_batch, sample_batched in enumerate(self.train_data_loader): #batch级别
global_step +=1
self.model.train() #进入训练状态
optimizer.zero_grad() #清除梯度
inputs = [sample_batched[col].to(self.opt.device) if col != 'context' else sample_batched[col] for col in self.opt.inputs_cols]
targets = sample_batched['label'].to(self.opt.device) #label
outputs = self.model(inputs)
loss=loss_funciton(outputs,targets)
loss.backward() #计算梯度
optimizer.step() #更新梯度
if global_step % self.opt.log_step==0: #log_step个batch训练后，测试一次性能
n_correct += (torch.argmax(outputs, -1) == targets).sum().item()
n_total += len(outputs)
train_acc=n_correct/n_total #训练集的准确率
test_acc, test_f1, test_precision, test_recall = self._evaluate_acc_f1() #测试
if test_f1 > max_test_f1: #f1是核心指标，如果有提升，就进来
max_test_acc=test_acc
max_test_recall=test_recall
max_test_precision=test_precision
increase_flag=True #性能有提升标志置为真
max_test_f1=test_f1
if self.opt.save and test_f1 > self.global_f1:
self.global_f1=test_f1
#在这保存模型？
#输出训练信息已经达到的最好的性能。
print("max test f1: ",max_test_f1,"max test acc: ",max_test_acc,"max test pre: ",max_test_precision,"max test recall: ",max_test_recall)
print('loss: {:.4f}, acc: {:.4f}, test_acc: {:.4f}, test_f1: {:.4f}'.format(loss.item(), train_acc, test_acc, test_f1))
if increase_flag ==False: #早停机制，如果在一个epoch里面，性能没有提升过。
continue_not_increase +=1
if continue_not_increase >= self.opt.estop: #忍耐度默认为4
print('early stop.')
break
else:
continue_not_increase=0 #如果有提升，也就是increase_flag为真，计数重置
return max_test_acc, max_test_f1 , max_test_precision , max_test_recall
def _evaluate_acc_f1(self):
'''
模型性能性能评估
:return:
'''
self.model.eval() #switch model to evaluation model
n_test_correct, n_test_total = 0, 0
t_targets_all, t_outputs_all = None, None
with torch.no_grad():
for t_batch, t_sample_batched in enumerate(self.test_data_loader):
t_inputs = [t_sample_batched[col].to(self.opt.device) if col != 'context' else t_sample_batched[col] for col
in self.opt.inputs_cols]
t_targets = t_sample_batched['label'].to(self.opt.device)
t_outputs = self.model(t_inputs)
n_test_correct += (torch.argmax(t_outputs, -1) == t_targets).sum().item()
n_test_total += len(t_outputs)
if t_targets_all is None:
t_targets_all = t_targets
t_outputs_all = t_outputs
else:
t_targets_all = torch.cat((t_targets_all, t_targets), dim=0)
t_outputs_all = torch.cat((t_outputs_all, t_outputs), dim=0)
test_acc = n_test_correct / n_test_total
f1 = metrics.f1_score(t_targets_all.cpu(), torch.argmax(t_outputs_all, -1).cpu(), labels=[0, 1],
average='macro')
precision_macro = metrics.precision_score(t_targets_all.cpu(), torch.argmax(t_outputs_all, -1).cpu(),
labels=[0, 1], average='macro')
recall_macro = metrics.recall_score(t_targets_all.cpu(), torch.argmax(t_outputs_all, -1).cpu(), labels=[0, 1],
average='macro')
return test_acc, f1, precision_macro, recall_macro
def _reset_params(self):
'''
给训练参数初始化
:return:
'''
for p in self.model.parameters():
if p.requires_grad:
if len(p.shape) > 1:
self.opt.initializer(p) #参数初始化，初始化方法由自己指定
else:
stdv = 1. / math.sqrt(p.shape[0])
torch.nn.init.uniform_(p, a=-stdv, b=stdv)
def main():
parser=argparse.ArgumentParser()
#需要的参数
parser.add_argument("--dataset",default="Semeval",help="数据集Semeval ")
parser.add_argument("--sentiment", default="SOLID", help="辅助任务的数据集名字 ")
parser.add_argument("--output_dir",default="./Output/",type=str,help="输出目录")
parser.add_argument("--pretrain_dir",default="./Pre-trained/",help="预训练模型所在的目录")
parser.add_argument('--optimizer', default='adam', type=str,help="优化器")
parser.add_argument("--max_seq_len",default=128,type=int,help="输入句子长度，默认128。")
parser.add_argument("--do_train",action='store_true',help="决定是否要训练")
parser.add_argument("--do_eval",action='store_true',help="决定是否要进行测试")
parser.add_argument("--learning_rate",default=5e-5,type=float,help="学习率")
parser.add_argument("--num_train_epochs",default=3.0,type=float,help="训练次数")
parser.add_argument('--device', default="cuda:0", type=str,help="训练设备")
parser.add_argument('--model_name', default='sksmt', type=str,help="要训练的模型名称") ###
parser.add_argument('--embed_dim', default=768, type=int,help="词向量维度")
parser.add_argument('--batch_size', default=16, type=int,help="batch大小")
parser.add_argument('--hidden_dim', default=768, type=int, help="LSTM的Hidden维度")
parser.add_argument('--polarities_dim', default=2, type=int,help="分类极性")
parser.add_argument('--repeat', default=3, type=int,help="训练几次")
parser.add_argument('--l2reg', default=0.00001, type=float,help="正则化")
parser.add_argument('--initializer', default='xavier_uniform_', type=str,help="训练参数初始化")
parser.add_argument('--num_epoch', default=100, type=int,help="迭代次数")
parser.add_argument('--log_step', default=5, type=int,help="几个batch更新一次")
parser.add_argument('--save', default=True, type=bool,help="是否保存最后的全局性能")
parser.add_argument('--estop', default=4, type=int,help="早停机制,4次性能没有进步就停止训练")
parser.add_argument('--log_info', default="运行完毕", type=str)
opt=parser.parse_args();
'''
为了方便Debug，配置数值会在这里改
'''
#####################################################测试结尾
optimizers = {#优化器
'adadelta': torch.optim.Adadelta, # default lr=1.0
'adagrad': torch.optim.Adagrad, # default lr=0.01
'adam': torch.optim.Adam, # default lr=0.001
'adamax': torch.optim.Adamax, # default lr=0.002
'asgd': torch.optim.ASGD, # default lr=0.01
'rmsprop': torch.optim.RMSprop, # default lr=0.01
'sgd': torch.optim.SGD,
}
model_classes = { #模型种类
'affgcn': AFFGCN,
'sksmt': SKSMT,
}
#数据输入形式
input_colses = {
'affgcn': ['context_indices', 'dependency_graph', 'sentiment_graph'],
'sksmt': ['context_indices','attention_mask','dependency_graph','sentiment_graph'],
}
#输入参数初始化
initializers = {
'xavier_uniform_': torch.nn.init.xavier_uniform_, #
'xavier_normal_': torch.nn.init.xavier_normal, #
'orthogonal_': torch.nn.init.orthogonal_, #
}
#用日志包，得到想要的信息
#日记包的基础配置
logging.basicConfig(filename="ModelLog.txt",filemode="a+",
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
level=logging.INFO,
datefmt='%m/%d/%Y %H:%M:%S'
)
opt.model_class =model_classes[opt.model_name]
opt.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # 如果GPU可用，用GPU
opt.inputs_cols = input_colses[opt.model_name]
opt.initializer = initializers[opt.initializer]
opt.optimizer = optimizers[opt.optimizer] #优化器选择
ins=Instructor(opt)
ins.run(opt.repeat)
if __name__ == '__main__':
main()