/
eval_cifar.py
152 lines (111 loc) · 4.34 KB
/
eval_cifar.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
import torch
from torch import nn
import numpy as np
import torch.optim as optim
def eval_sgd(x_train, y_train, x_test, y_test, topk=[1, 5], epoch=500, batch_size=1000):
""" linear classifier accuracy (sgd) """
lr_start, lr_end = 1e-2, 1e-6
gamma = (lr_end / lr_start) ** (1 / epoch)
output_size = x_train.shape[1]
num_class = y_train.max().item() + 1
clf = nn.Linear(output_size, num_class)
clf.cuda()
clf.train()
optimizer = optim.Adam(clf.parameters(), lr=lr_start, weight_decay=5e-6)
scheduler = optim.lr_scheduler.ExponentialLR(optimizer, gamma=gamma)
criterion = nn.CrossEntropyLoss()
for ep in range(epoch):
perm = torch.randperm(len(x_train))
n_batch = int(np.ceil(len(x_train)*1.0/batch_size))
for ii in range(n_batch):
optimizer.zero_grad()
mask = perm[ii*batch_size:(ii+1)*batch_size]
criterion(clf(x_train[mask]), y_train[mask]).backward()
optimizer.step()
scheduler.step()
clf.eval()
with torch.no_grad():
y_pred = clf(x_test)
pred_top = y_pred.topk(max(topk), 1, largest=True, sorted=True).indices
acc = {
t: (pred_top[:, :t] == y_test[..., None]).float().sum(1).mean().cpu().item()
for t in topk
}
del clf
return acc
def eval_sgd_per_class(x_train, y_train, x_test, y_test, topk=[1, 5], epoch=500, batch_size=1000):
num_class = y_train.max().item() + 1
perClassAccRight = [0 for _ in range(num_class)]
perClassAccWhole = [0 for _ in range(num_class)]
perClassAcc = [0 for _ in range(num_class)]
""" linear classifier accuracy (sgd) """
lr_start, lr_end = 1e-2, 1e-6
gamma = (lr_end / lr_start) ** (1 / epoch)
output_size = x_train.shape[1]
clf = nn.Linear(output_size, num_class)
clf.cuda()
clf.train()
optimizer = optim.Adam(clf.parameters(), lr=lr_start, weight_decay=5e-6)
scheduler = optim.lr_scheduler.ExponentialLR(optimizer, gamma=gamma)
criterion = nn.CrossEntropyLoss()
for ep in range(epoch):
perm = torch.randperm(len(x_train))
n_batch = int(np.ceil(len(x_train)*1.0/batch_size))
for ii in range(n_batch):
optimizer.zero_grad()
mask = perm[ii*batch_size:(ii+1)*batch_size]
criterion(clf(x_train[mask]), y_train[mask]).backward()
optimizer.step()
scheduler.step()
clf.eval()
with torch.no_grad():
y_pred = clf(x_test)
pred_top = y_pred.topk(max(topk), 1, largest=True, sorted=True).indices
acc = {
t: (pred_top[:, :t] == y_test[..., None]).float().sum(1).mean().cpu().item()
for t in topk
}
y_pred = y_pred.max(1, keepdim=True)[1]
for i in range(num_class):
perClassAccRight[i] = y_pred[y_test == i].eq(y_test[y_test == i].view_as(y_pred[y_test == i])).sum().item()
perClassAccWhole[i] = len(y_test[y_test == i])
for i in range(num_class):
perClassAcc[i] = perClassAccRight[i] / perClassAccWhole[i] * 100
del clf
return acc, perClassAcc
def eval(train_loader, test_loader, model, epoch, args=None):
model.eval()
if args.prune_percent > 0:
# sdclr, BCL-D
projector = model.backbone.fc
model.backbone.fc = nn.Identity()
else:
# simclr, BCL-I
projector = model.projector
model.projector = nn.Identity()
with torch.no_grad():
model.eval()
x_train = []
y_train = []
x_test = []
y_test = []
for i, (inputs, labels) in enumerate(train_loader):
inputs = inputs.cuda()
features = model(inputs)
x_train.append(features.detach())
y_train.append(labels.detach())
for i, (inputs, labels) in enumerate(test_loader):
inputs = inputs.cuda()
features = model(inputs)
x_test.append(features.detach())
y_test.append(labels.detach())
x_train = torch.cat(x_train, dim=0)
y_train = torch.cat(y_train, dim=0).cuda()
x_test = torch.cat(x_test, dim=0)
y_test = torch.cat(y_test, dim=0).cuda()
acc = eval_sgd(x_train, y_train, x_test, y_test)
if args.prune_percent > 0:
model.backbone.fc = projector
else:
model.projector = projector
return acc