-
Notifications
You must be signed in to change notification settings - Fork 0
/
trainer.py
128 lines (86 loc) · 4.06 KB
/
trainer.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
import torch
from torch.autograd import Variable
import pickle
def read(path):
with open(path, 'rb') as f:
data = pickle.load(f)
return data
def train(config, model, train_loader, val_loader, optimizer, criterion):
#TA, VA = [], []
#TL, VL = [], []
TA = read('results/train_accuracy.pkl')
VA = read('results/val_accuracy.pkl')
TL = read('results/train_loss.pkl')
VL = read('results/val_loss.pkl')
for epoch in range(config['epochs']):
model.train()
loss_train = 0
total, correct = 0, 0
for i, (x1, x2, y) in enumerate(train_loader):
if torch.cuda.is_available():
x1 = x1.cuda()
x2 = x2.cuda()
y = y.cuda()
x1 = Variable(x1)
x2 = Variable(x2)
y = Variable(y)
optimizer.zero_grad()
outputs = model(x1, x2)
train_loss = criterion(outputs, y)
_, predicted = torch.max(outputs.data, 1)
total += y.size(0)
if torch.cuda.is_available():
correct += (y.cpu() == predicted.cpu()).sum(0)
else:
correct += (y == predicted).sum(0)
loss_train += train_loss.item()*x1.size(0)
train_loss.backward()
optimizer.step()
if i%100 == 0:
print('[%d/%d] \t Train Loss: %.4f \t Train Acc: %.4f \t %d/%d' % (i,
len(train_loader),
loss_train/(i+1),
100.0*float(correct.item())/total,
correct,
total))
torch.save(model, 'checkpoint/model_vgg16.pth')
model.eval()
loss_val = 0
total_val, correct_val = 0, 0
for i, (x1, x2, y) in enumerate(val_loader):
if torch.cuda.is_available():
x1 = x1.cuda()
x2 = x2.cuda()
y = y.cuda()
x1 = Variable(x1)
x2 = Variable(x2)
y = Variable(y)
outputs = model(x1, x2)
val_loss = criterion(outputs, y)
_, predicted = torch.max(outputs.data, 1)
total_val += y.size(0)
if torch.cuda.is_available():
correct_val += (y.cpu() == predicted.cpu()).sum(0)
else:
correct_val += (y == predicted).sum(0)
loss_val += val_loss.item()*x1.size(0)
print('-'*120)
print('Epoch %d \t Train Loss: %.3f \t Val Loss: %.3f \t Train Acc: %.3f \t Val Acc: %.3f'% ( epoch+1,
loss_train/len(train_loader),
loss_val/len(val_loader),
100.0*correct.item()/total,
100.0*correct_val.item()/total_val ))
TL.append(loss_train/len(train_loader))
VL.append(loss_val/len(val_loader))
TA.append(100.0*correct.item()/total)
VA.append(100.0*correct_val.item()/total_val)
with open('results/train_accuracy.pkl', 'wb') as f:
pickle.dump(TA, f)
with open('results/val_accuracy.pkl', 'wb') as f:
pickle.dump(VA, f)
with open('results/train_loss.pkl', 'wb') as f:
pickle.dump(TL, f)
with open('results/val_loss.pkl', 'wb') as f:
pickle.dump(VL, f)
print('-'*120)
print('\n')