/
scl.py
186 lines (128 loc) · 6.34 KB
/
scl.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
import torch
from args import args
def train_epoch_scl(encoder, projector, train_loader, train_transform, criterion, optimizer, scheduler):
epoch_loss = 0.0
for data, target, _ in train_loader:
data, target = data.to(args.device), target.to(args.device)
with torch.no_grad():
data_t1 = train_transform(data)
data_t2 = train_transform(data)
feat1, feat2 = encoder(data_t1), encoder(data_t2)
proj1, proj2 = projector(feat1), projector(feat2)
loss = criterion(proj1, proj2, target)
epoch_loss += loss.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
epoch_loss = epoch_loss / len(train_loader)
scheduler.step()
return epoch_loss
def train_scl(encoder, projector, train_loader, train_transform, criterion, optimizer, scheduler, epochs):
best_loss = None
train_losses = []
encoder.train()
projector.train()
for i in range(1, epochs+1):
print(f"Epoch {i}")
train_loss = train_epoch_scl(encoder, projector, train_loader, train_transform, criterion, optimizer, scheduler)
print(f"Current Train Loss : {format(train_loss, '.4f')}")
train_losses.append(train_loss)
if best_loss is None:
best_loss = train_loss
if best_loss > train_loss:
best_loss = train_loss
final_state = {"encoder": encoder.state_dict()}
print(f"Last Loss : {format(train_loss, '.4f')}\tBest Loss : {format(best_loss, '.4f')}")
return train_losses, encoder, final_state
def linear_train_epoch(encoder, classifier, train_loader, val_transform, criterion, optimizer):
epoch_loss = 0.0
if args.dataset == 'ICBHI':
TP = [0, 0, 0 ,0]
GT = [0, 0, 0, 0]
elif args.dataset == 'SPRS':
TP = [0, 0, 0 ,0, 0, 0, 0]
GT = [0, 0, 0, 0, 0, 0, 0]
classifier.train()
for data, target, _ in train_loader:
data, target = data.to(args.device), target.to(args.device)
with torch.no_grad():
features = encoder(val_transform(data))
optimizer.zero_grad()
output = classifier(features)
loss = criterion(output, target)
epoch_loss += loss.item()
_, labels_predicted = torch.max(output, dim=1)
for idx in range(len(TP)):
TP[idx] += torch.logical_and((labels_predicted==idx),(target==idx)).sum().item()
GT[idx] += (target==idx).sum().item()
loss.backward()
optimizer.step()
epoch_loss = epoch_loss / len(train_loader)
se = sum(TP[1:])/sum(GT[1:])
sp = TP[0]/GT[0]
icbhi_score = (se+sp)/2
acc = sum(TP)/sum(GT)
return epoch_loss, se, sp, icbhi_score, acc
def linear_eval_epoch(encoder, classifier, val_loader, val_transform, criterion):
epoch_loss = 0.0
if args.dataset == 'ICBHI':
TP = [0, 0, 0 ,0]
GT = [0, 0, 0, 0]
elif args.dataset == 'SPRS':
TP = [0, 0, 0 ,0, 0, 0, 0]
GT = [0, 0, 0, 0, 0, 0, 0]
classifier.eval()
encoder.eval()
with torch.no_grad():
for data, target, _ in val_loader:
data, target = data.to(args.device), target.to(args.device)
output = classifier(encoder(val_transform(data)))
loss = criterion(output, target)
epoch_loss += loss.item()
_, labels_predicted = torch.max(output, dim=1)
for idx in range(len(TP)):
TP[idx] += torch.logical_and((labels_predicted==idx),(target==idx)).sum().item()
GT[idx] += (target==idx).sum().item()
epoch_loss = epoch_loss / len(val_loader)
se = sum(TP[1:])/sum(GT[1:])
sp = TP[0]/GT[0]
icbhi_score = (se+sp)/2
acc = sum(TP)/sum(GT)
return epoch_loss, se, sp, icbhi_score, acc
def linear_scl(encoder, checkpoint, classifier, train_loader, val_loader, val_transform, criterion, optimizer, epochs):
train_losses = []; val_losses = []; train_se_scores = []; train_sp_scores = []; train_icbhi_scores = []; train_acc_scores = []; val_se_scores = []; val_sp_scores = []; val_icbhi_scores = []; val_acc_scores = []
best_val_acc = 0
best_icbhi_score = 0
best_se = 0
best_sp = 0
best_epoch_acc = 0
best_epoch_icbhi = 0
state_dict = checkpoint["encoder"]
encoder.load_state_dict(state_dict)
for param in encoder.parameters():
param.requires_grad = False
encoder.eval()
for i in range(1, epochs+1):
print(f"Epoch {i}")
train_loss, train_se, train_sp, train_icbhi_score, train_acc = linear_train_epoch(encoder, classifier, train_loader, val_transform, criterion, optimizer)
train_losses.append(train_loss); train_se_scores.append(train_se); train_sp_scores.append(train_sp); train_icbhi_scores.append(train_icbhi_score); train_acc_scores.append(train_acc)
print(f"Train loss : {format(train_loss, '.4f')}\tTrain SE : {format(train_se, '.4f')}\tTrain SP : {format(train_sp, '.4f')}\tTrain Score : {format(train_icbhi_score, '.4f')}\tTrain Acc : {format(train_acc, '.4f')}")
val_loss, val_se, val_sp, val_icbhi_score, val_acc = linear_eval_epoch(encoder, classifier, val_loader, val_transform, criterion)
val_losses.append(val_loss); val_se_scores.append(val_se); val_sp_scores.append(val_sp); val_icbhi_scores.append(val_icbhi_score); val_acc_scores.append(val_acc)
print(f"Val loss : {format(val_loss, '.4f')}\tVal SE : {format(val_se, '.4f')}\tVal SP : {format(val_sp, '.4f')}\tVal Score : {format(val_icbhi_score, '.4f')}\tVal Acc : {format(val_acc, '.4f')}")
if best_val_acc == 0:
best_val_acc = val_acc
if i == 1:
best_icbhi_score = val_icbhi_score
best_se = val_se
best_sp = val_sp
if best_icbhi_score < val_icbhi_score:
best_epoch_icbhi = i
best_icbhi_score = val_icbhi_score
best_se = val_se
best_sp = val_sp
if best_val_acc < val_acc:
best_epoch_acc = i
best_val_acc = val_acc
print(f"best score is {format(best_icbhi_score, '.4f')} (se:{format(best_se, '.4f')} sp:{format(best_sp, '.4f')}) at epoch {best_epoch_icbhi}")
return train_losses, val_losses, train_se_scores, train_sp_scores, train_icbhi_scores, train_acc_scores, val_se_scores, val_sp_scores, val_icbhi_scores, val_acc_scores