-
Notifications
You must be signed in to change notification settings - Fork 1
/
linear_binary.py
161 lines (129 loc) · 7.08 KB
/
linear_binary.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 argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from tqdm import tqdm
import sys
import os
import shutil
import pickle
import logging
from logger import *
import utils
from model import Model, Image_Model
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print('Home device: {}'.format(device))
class Net(nn.Module):
def __init__(self, num_class, pretrained_path):
super(Net, self).__init__()
# encoder
model = Image_Model().to(device)
model = nn.DataParallel(model)
model.load_state_dict(torch.load(pretrained_path, map_location='cuda:0'))
# self.f=model.f
self.f = model.module.model
# classifier
self.fc = nn.Linear(512, num_class, bias=True)
def forward(self, x):
x = self.f(x)
feature = torch.flatten(x, start_dim=1)
out = self.fc(feature)
return out
# train or test for one epoch
def train_val_binary(net, data_loader, train_optimizer):
is_train = train_optimizer is not None
net.train() if is_train else net.eval()
num_class = len(data_loader.dataset.classes)
num_sens = len(data_loader.dataset.sens_groups)
num_group = num_sens * num_class
total_loss, total_correct, total_num, data_bar = 0.0, 0.0, 0, tqdm(data_loader)
total_correct_min = np.array([0] * num_group)
total_num_class = np.array([0] * num_group)
with (torch.enable_grad() if is_train else torch.no_grad()):
for data, sens, target in data_bar:
data, sens, target = data.to(device,non_blocking=True), sens.to(device,non_blocking=True), target.to(device,non_blocking=True)
out = net(data)
loss = loss_criterion(out, target)
if is_train:
train_optimizer.zero_grad()
loss.backward()
train_optimizer.step()
total_num += data.size(0)
total_loss += loss.item() * data.size(0)
prediction = torch.argsort(out, dim=-1, descending=True)
total_correct += torch.sum((prediction[:, 0:1] == target.unsqueeze(dim=-1)).any(dim=-1).float()).item()
group = abs(target) * 2 + abs(sens)
for i in range(num_group):
idx = group == i
total_correct_min[i] += torch.sum((prediction[idx, 0:1] == target[idx].unsqueeze(dim=-1)).any(dim=-1).float()).item()
total_num_class[i] += sum(idx).item()
desc = '{} Epoch: [{}/{}] Loss: {:.4f} ACC: {:.2f}% ||'\
.format('Train' if is_train else 'Test', epoch, epochs, total_loss / total_num,\
total_correct / total_num * 100)
for i in range(num_group):
desc += 'group {} acc : {:.2f}%, '.format(i, (total_correct_min[i] / total_num_class[i]) * 100)
data_bar.set_description(desc)
logger_train.info(desc)
return total_loss / total_num, total_correct / total_num * 100, min(total_correct_min / total_num_class) * 100, np.argmin(total_correct_min / total_num_class), \
[(total_correct_min[i] / total_num_class[i]) * 100 for i in range(num_group)]
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Linear Evaluation')
parser.add_argument('--root', type=str, default='../data', help='Path to data directory')
parser.add_argument('--model_path', type=str, default='../results/cifar10/wcl/no_orient/cifar10_model_b_256_0.1_0.001_250.pth',
help='The pretrained model path')
parser.add_argument('--batch_size', type=int, default=256, help='Number of images in each mini-batch')
parser.add_argument('--epochs', type=int, default=100, help='Number of sweeps over the dataset to train')
parser.add_argument('--dataset_name', default='cifar10', type=str, help='Choose loss function')
parser.add_argument('--model_name', default='wcl', type=str, help='Choose model name')
parser.add_argument('--lr', default=1e-3, type=float, help='learning rate')
args = parser.parse_args()
model_path, batch_size, epochs = args.model_path, args.batch_size, args.epochs
dataset_name = args.dataset_name
lr = args.lr
save_dir = os.path.split(model_path)[0]
model_config = os.path.split(model_path)[1]
# save_dir = '../results/{}'.format(args.dataset_name)
if not os.path.exists(save_dir):
os.makedirs(save_dir)
# train_logger = Logger(train_log_dir)
setup_logger('log_train', os.path.join(save_dir, 'log_cls_b_{}.txt'.format(model_config)))
logger_train = logging.getLogger('log_train')
shutil.copyfile('linear_binary.py', os.path.join(save_dir, 'linear_binary.py'))
logger_train.info(sys.argv)
save_pkl = os.path.join(save_dir, 'linear_b_summary_{}.pkl'.format(args.model_name))
train_data, _, test_data = utils.get_dataset(dataset_name, root=args.root, pair=False)
train_loader = DataLoader(train_data, batch_size=batch_size, shuffle=True, num_workers=8, pin_memory=True)
test_loader = DataLoader(test_data, batch_size=batch_size, shuffle=False, num_workers=8, pin_memory=True)
model = Net(num_class=len(train_data.classes), pretrained_path=model_path).to(device)
for param in model.f.parameters():
param.requires_grad = False
model = nn.DataParallel(model)
optimizer = optim.SGD(model.module.fc.parameters(), lr=lr, weight_decay=1e-4, momentum = 0.9)
loss_criterion = nn.CrossEntropyLoss()
results = {'train_loss': [], 'train_acc': [], 'train_acc_group': [], 'train_acc(min)' : [], 'worst_group':[],
'test_loss': [], 'test_acc': [], 'test_acc_group': [], 'test_acc(min)' : [], 'worst_group(test)':[]}
for epoch in range(1, epochs + 1):
train_loss, train_acc, train_acc_min, worst_group, group_acc = train_val_binary(model, train_loader, optimizer)
results['train_loss'].append(train_loss)
results['train_acc'].append(train_acc)
results['train_acc_group'].append(group_acc)
results['train_acc(min)'].append(train_acc_min)
results['worst_group'].append(worst_group)
if epoch % 5 == 0:
test_loss, test_acc, test_acc_min, worst_group, group_acc = train_val_binary(model, test_loader, None)
# os.makedirs('../results/')
# try:
# results=pickle.load( open(save_dir, "rb" ))
# except:
# results = {'train_loss': [], 'train_acc@1': [], 'train_acc@5': [], 'train_acc(min)' : [],
# 'test_loss': [], 'test_acc@1': [], 'test_acc@5': [], 'test_acc(min)' : []}
# results[model_path]=test_acc_1
results['test_loss'].append(test_loss)
results['test_acc'].append(test_acc)
results['test_acc_group'].append(group_acc)
results['test_acc(min)'].append(test_acc_min)
results['worst_group(test)'].append(worst_group)
pickle.dump(results, open( save_pkl, "wb" ) )
torch.save(model.state_dict(), os.path.join(save_dir, 'summary_{}.pth'.format(model_config)))