-
Notifications
You must be signed in to change notification settings - Fork 4
/
train_test_helper.py
161 lines (114 loc) · 5.61 KB
/
train_test_helper.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
import torch
import torch.nn.functional as F
from torch.autograd import Variable
from torch.nn.utils import clip_grad_norm_
def get_count_correct_preds(network_output, target):
output = network_output
pred = output.data.max(1, keepdim=True)[1] # get the index of the max log-probability
pred.data = pred.data.view_as(target.data)
correct = target.eq(pred).sum().item()
return correct
class ModelTrainTest():
def __init__(self, network, device, model_file_path, threshold=1e-4):
super(ModelTrainTest, self).__init__()
self.network = network
self.device = device
self.model_file_path = model_file_path
self.threshold = threshold
self.train_loss = 1e9
self.val_loss = 1e9
def train(self, optimizer, epoch, params_max_norm, train_data_loader, val_data_loader):
self.network.train()
train_loss = 0
correct = 0
cnt_batches = 0
for batch_idx, (data, target) in enumerate(train_data_loader):
data, target = Variable(data).to(self.device), Variable(target).to(self.device)
optimizer.zero_grad()
output = self.network(data)
loss = F.nll_loss(output, target)
loss.backward()
clip_grad_norm_(self.network.parameters(), params_max_norm)
optimizer.step()
correct += get_count_correct_preds(output, target)
train_loss += loss.item()
cnt_batches += 1
del data, target, output
train_loss /= cnt_batches
val_loss, val_acc = self.test(epoch, val_data_loader)
if val_loss < self.val_loss - self.threshold:
self.val_loss = val_loss
torch.save(self.network.state_dict(), self.model_file_path)
train_acc = correct / len(train_data_loader.dataset)
print('\nAfter epoch {} - Train set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
epoch, train_loss, correct, len(train_data_loader.dataset),
100. * correct / len(train_data_loader.dataset)))
return train_loss, train_acc, val_loss, val_acc
def test(self, epoch, test_data_loader):
self.network.eval()
test_loss = 0
correct = 0
for batch_idx, (data, target) in enumerate(test_data_loader):
data, target = Variable(data, volatile=True).to(self.device), Variable(target).to(self.device)
output = self.network(data)
test_loss += F.nll_loss(output, target, size_average=False).item() # sum up batch loss
correct += get_count_correct_preds(output, target)
del data, target, output
test_loss /= len(test_data_loader.dataset)
test_acc = correct / len(test_data_loader.dataset)
print('\nAfter epoch {} - Test set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
epoch, test_loss, correct, len(test_data_loader.dataset),
100. * correct / len(test_data_loader.dataset)))
return test_loss, test_acc
class JigsawModelTrainTest():
def __init__(self, network, device, model_file_path, threshold=1e-4):
super(JigsawModelTrainTest, self).__init__()
self.network = network
self.device = device
self.model_file_path = model_file_path
self.threshold = threshold
self.train_loss = 1e9
self.val_loss = 1e9
def train(self, optimizer, epoch, params_max_norm, train_data_loader, val_data_loader):
self.network.train()
train_loss = 0
correct = 0
cnt_batches = 0
for batch_idx, (data, target) in enumerate(train_data_loader):
data, target = Variable(data).to(self.device), Variable(target).to(self.device)
optimizer.zero_grad()
output = self.network(data)
loss = F.nll_loss(output, target)
loss.backward()
clip_grad_norm_(self.network.parameters(), params_max_norm)
optimizer.step()
correct += get_count_correct_preds(output, target)
train_loss += loss.item()
cnt_batches += 1
del data, target, output
train_loss /= cnt_batches
val_loss, val_acc = self.test(epoch, val_data_loader)
if val_loss < self.val_loss - self.threshold:
self.val_loss = val_loss
torch.save(self.network.state_dict(), self.model_file_path)
train_acc = correct / len(train_data_loader.dataset)
print('\nAfter epoch {} - Train set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
epoch, train_loss, correct, len(train_data_loader.dataset),
100. * correct / len(train_data_loader.dataset)))
return train_loss, train_acc, val_loss, val_acc
def test(self, epoch, test_data_loader):
self.network.eval()
test_loss = 0
correct = 0
for batch_idx, (data, target) in enumerate(test_data_loader):
data, target = Variable(data, volatile=True).to(self.device), Variable(target).to(self.device)
output = self.network(data)
test_loss += F.nll_loss(output, target, size_average=False).item() # sum up batch loss
correct += get_count_correct_preds(output, target)
del data, target, output
test_loss /= len(test_data_loader.dataset)
test_acc = correct / len(test_data_loader.dataset)
print('\nAfter epoch {} - Test set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
epoch, test_loss, correct, len(test_data_loader.dataset),
100. * correct / len(test_data_loader.dataset)))
return test_loss, test_acc