-
Notifications
You must be signed in to change notification settings - Fork 76
/
Copy pathmain.py
201 lines (165 loc) · 5.21 KB
/
main.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
#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
@Description: : main function to carry out training and testing
@Author : Kevinpro
@version : 1.0
'''
import numpy as np
from sklearn.preprocessing import MinMaxScaler
import time
import copy
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.autograd as autograd
import torch.nn.functional
from torch.utils.data import Dataset, DataLoader
from transformers import AdamW
import warnings
import torch
import time
import argparse
import json
import os
from transformers import BertTokenizer
from transformers import BertForSequenceClassification
#from transformers import BertPreTrainedModel
from transformers import BertModel
from loader import load_train
from loader import load_dev
from loader import map_id_rel
import random
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
setup_seed(44)
rel2id, id2rel = map_id_rel()
print(len(rel2id))
print(id2rel)
def get_model():
labels_num=len(rel2id)
from model import BERT_Classifier
model = BERT_Classifier(labels_num)
return model
model=get_model()
# torch.save(model, './bert-base-chinese/test'+'.pth')
# exit()
# exit()
USE_CUDA = torch.cuda.is_available()
#USE_CUDA=False
data=load_train()
train_text=data['text']
train_mask=data['mask']
train_label=data['label']
train_text = [ t.numpy() for t in train_text]
train_mask = [ t.numpy() for t in train_mask]
train_text=torch.tensor(train_text)
train_mask=torch.tensor(train_mask)
train_label=torch.tensor(train_label)
print("--train data--")
print(train_text.shape)
print(train_mask.shape)
print(train_label.shape)
data=load_dev()
dev_text=data['text']
dev_mask=data['mask']
dev_label=data['label']
dev_text = [ t.numpy() for t in dev_text]
dev_mask = [ t.numpy() for t in dev_mask]
dev_text=torch.tensor(dev_text)
dev_mask=torch.tensor(dev_mask)
dev_label=torch.tensor(dev_label)
print("--train data--")
print(train_text.shape)
print(train_mask.shape)
print(train_label.shape)
print("--eval data--")
print(dev_text.shape)
print(dev_mask.shape)
print(dev_label.shape)
# exit()
#USE_CUDA=False
if USE_CUDA:
print("using GPU")
train_dataset = torch.utils.data.TensorDataset(train_text,train_mask,train_label)
dev_dataset = torch.utils.data.TensorDataset(dev_text,dev_mask,dev_label)
def eval(net,dataset, batch_size):
net.eval()
train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=False)
with torch.no_grad():
correct = 0
total=0
iter = 0
for text,mask, y in train_iter:
iter += 1
if text.size(0)!=batch_size:
break
text=text.reshape(batch_size,-1)
mask = mask.reshape(batch_size, -1)
if USE_CUDA:
text=text.cuda()
mask=mask.cuda()
y=y.cuda()
outputs= net(text, mask,y)
#print(y)
loss, logits = outputs[0],outputs[1]
_, predicted = torch.max(logits.data, 1)
total += text.size(0)
correct += predicted.data.eq(y.data).cpu().sum()
s = ("Acc:%.3f" %((1.0*correct.numpy())/total))
acc= (1.0*correct.numpy())/total
print("Eval Result: right", correct.cpu().numpy().tolist(), "total", total, "Acc:", acc)
return acc
def train(net,dataset,num_epochs, learning_rate, batch_size):
net.train()
optimizer = optim.SGD(net.parameters(), lr=learning_rate,weight_decay=0)
#optimizer = AdamW(net.parameters(), lr=learning_rate)
train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=True)
pre=0
for epoch in range(num_epochs):
correct = 0
total=0
iter = 0
for text,mask, y in train_iter:
iter += 1
optimizer.zero_grad()
#print(type(y))
#print(y)
if text.size(0)!=batch_size:
break
text=text.reshape(batch_size,-1)
mask = mask.reshape(batch_size, -1)
if USE_CUDA:
text=text.cuda()
mask=mask.cuda()
y = y.cuda()
#print(text.shape)
loss, logits= net(text, mask,y)
#print(y)
#print(loss.shape)
#print("predicted",predicted)
#print("answer", y)
loss.backward()
optimizer.step()
#print(outputs[1].shape)
#print(output)
#print(outputs[1])
_, predicted = torch.max(logits.data, 1)
total += text.size(0)
correct += predicted.data.eq(y.data).cpu().sum()
loss = loss.detach().cpu()
print("epoch ", str(epoch)," loss: ", loss.mean().numpy().tolist(),"right", correct.cpu().numpy().tolist(), "total", total, "Acc:", correct.cpu().numpy().tolist()/total)
acc = eval(model, dev_dataset, 32)
if acc > pre:
pre = acc
torch.save(model, str(acc)+'.pth')
return
#model=nn.DataParallel(model,device_ids=[0,1])
if USE_CUDA:
model=model.cuda()
#eval(model,dev_dataset,8)
train(model,train_dataset,1,0.002,4)
#eval(model,dev_dataset,8)