-
Notifications
You must be signed in to change notification settings - Fork 13
/
train.py
147 lines (118 loc) · 6.03 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
import sys
import numpy as np
import torch
import torch.nn.functional as F
def co_guess(net, net2, inputs_x, inputs_u, inputs_x2, inputs_u2, w_x, labels_x, T, smooth_clean):
# label co-guessing of unlabeled samples
outputs_u11 = net(inputs_u)
outputs_u12 = net(inputs_u2)
outputs_u21 = net2(inputs_u)
outputs_u22 = net2(inputs_u2)
pu = (torch.softmax(outputs_u11, dim=1) +
torch.softmax(outputs_u12, dim=1) +
torch.softmax(outputs_u21, dim=1) +
torch.softmax(outputs_u22, dim=1)) / 4
ptu = pu ** (1 / T) # temperature sharpening
targets_u = ptu / ptu.sum(dim=1, keepdim=True) # normalize
targets_u = targets_u.detach()
# label refinement of labeled samples
outputs_x = net(inputs_x)
outputs_x2 = net(inputs_x2)
if smooth_clean:
px = (torch.softmax(outputs_x, dim=1) + torch.softmax(outputs_x2, dim=1)) / 2
px = w_x * labels_x + (1 - w_x) * px
ptx = px ** (1 / T) # temparature sharpening
targets_x = ptx / ptx.sum(dim=1, keepdim=True) # normalize
targets_x = targets_x.detach()
else:
targets_x = labels_x
return targets_x, targets_u
# Training
def train(epoch, net, net2, criterion, optimizer, labeled_trainloader, unlabeled_trainloader, lambda_u, batch_size,
num_class, device, T, alpha, warm_up, dataset, r, noise_mode, num_epochs, smooth_clean=True):
net.train()
net2.eval() # fix one network and train the other
unlabeled_train_iter = iter(unlabeled_trainloader)
num_iter = (len(labeled_trainloader.dataset) // batch_size) + 1
for batch_idx, (inputs_x, inputs_x2, labels_x, _, w_x) in enumerate(labeled_trainloader):
try:
inputs_u, inputs_u2 = unlabeled_train_iter.next()
except:
unlabeled_train_iter = iter(unlabeled_trainloader)
inputs_u, inputs_u2 = unlabeled_train_iter.next()
batch_size = inputs_x.size(0)
# Transform label to one-hot
labels_x = torch.zeros(batch_size, num_class).scatter_(1, labels_x.view(-1, 1), 1)
w_x = w_x.view(-1, 1).type(torch.FloatTensor)
inputs_x, inputs_x2 = inputs_x.to(device), inputs_x2.to(device)
labels_x, w_x = labels_x.to(device), w_x.to(device)
inputs_u, inputs_u2 = inputs_u.to(device), inputs_u2.to(device)
with torch.no_grad():
# label co-guessing of unlabeled samples
targets_x, targets_u = co_guess(net, net2, inputs_x, inputs_u, inputs_x2, inputs_u2, w_x, labels_x, T,
smooth_clean)
# mixmatch
l = np.random.beta(alpha, alpha)
l = max(l, 1 - l)
all_inputs = torch.cat([inputs_x, inputs_x2, inputs_u, inputs_u2], dim=0)
all_targets = torch.cat([targets_x, targets_x, targets_u, targets_u], dim=0)
idx = torch.randperm(all_inputs.size(0))
input_a, input_b = all_inputs, all_inputs[idx]
target_a, target_b = all_targets, all_targets[idx]
if lambda_u > 0:
mixed_input = l * input_a + (1 - l) * input_b
mixed_target = l * target_a + (1 - l) * target_b
logits = net(mixed_input)
logits_x = logits[:batch_size * 2]
logits_u = logits[batch_size * 2:]
Lx, Lu, lamb = criterion(logits_x, mixed_target[:batch_size * 2], logits_u, mixed_target[batch_size * 2:],
epoch + batch_idx / num_iter, warm_up, lambda_u)
else:
mixed_input = l * input_a[:batch_size * 2] + (1 - l) * input_b[:batch_size * 2]
mixed_target = l * target_a[:batch_size * 2] + (1 - l) * target_b[:batch_size * 2]
logits = net(mixed_input)
Lx = -torch.mean(torch.sum(F.log_softmax(logits, dim=1) * mixed_target, dim=1))
lamb, Lu = 0, 0
# regularization
prior = torch.ones(num_class) / num_class
prior = prior.to(device)
pred_mean = torch.softmax(logits, dim=1).mean(0)
penalty = torch.sum(prior * torch.log(prior / pred_mean))
loss = Lx + lamb * Lu + penalty
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
optimizer.step()
sys.stdout.write('\r')
if 'cifar' in 'dataset':
sys.stdout.write('%s:%.1f-%s | Epoch [%3d/%3d] Iter[%3d/%3d]\t'
'Labeled loss: %.2f Unlabeled loss: %.2e(%.2e) penalty: %.2e'
% (dataset, r, noise_mode, epoch, num_epochs, batch_idx + 1, num_iter,
Lx.item(), Lu.item(), lamb * Lu.item(), penalty.item()))
elif 'clothing' in dataset:
sys.stdout.write('Clothing1M | Epoch [%3d/%3d] Iter[%3d/%3d]\t'
'Labeled loss: %.2f penalty: %.2e'
% (epoch, num_epochs, batch_idx + 1, num_iter, Lx.item(), penalty.item()))
sys.stdout.flush()
def warmup(epoch, net, optimizer, dataloader, criterion, conf_penalty, device, dataset, r, num_epochs, noise_mode):
net.train()
for batch_idx, (inputs, _, labels, _, _) in enumerate(dataloader):
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, labels)
assert torch.isfinite(loss).all()
penalty = conf_penalty(outputs) if conf_penalty is not None else 0.
L = loss + penalty
L.backward()
torch.nn.utils.clip_grad_norm_(net.parameters(), 1.)
optimizer.step()
sys.stdout.write('\r')
if 'clothing' in dataset:
sys.stdout.write('|Warm-up: Iter[%3d/%3d]\t CE-loss: %.4f Conf-Penalty: %.4f'
% (batch_idx + 1, len(dataloader), loss.item(), penalty.item()))
elif 'cifar' in dataset:
sys.stdout.write('%s:%.1f-%s | Epoch [%3d/%3d] Iter[%3d/%3d]\t CE-loss: %.4f'
% (dataset, r, noise_mode, epoch, num_epochs, batch_idx + 1, len(dataloader),
loss.item()))
sys.stdout.flush()