-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_lora_jocor.py
175 lines (144 loc) · 6.69 KB
/
train_lora_jocor.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
import os
import torch
import torch.nn as nn
import argparse
import timm
import numpy as np
import utils
import random
import rein
import others.jocor
import time
import dino_variant
from sklearn.metrics import f1_score
def train():
parser = argparse.ArgumentParser()
parser.add_argument('--data', '-d', type=str)
parser.add_argument('--gpu', '-g', default = '0', type=str)
parser.add_argument('--netsize', default='s', type=str)
parser.add_argument('--save_path', '-s', type=str)
parser.add_argument('--noise_rate', '-n', type=float, default=0.2)
args = parser.parse_args()
config = utils.read_conf('conf/'+args.data+'.json')
device = 'cuda:'+args.gpu
save_path = os.path.join(config['save_path'], args.save_path)
data_path = config['id_dataset']
batch_size = int(config['batch_size'])
max_epoch = int(config['epoch'])
noise_rate = args.noise_rate
if not os.path.exists(save_path):
os.mkdir(save_path)
lr_decay = [int(0.5*max_epoch), int(0.75*max_epoch), int(0.9*max_epoch)]
num_gradual = 10
if args.data == 'ham10000':
train_loader, valid_loader = utils.get_noise_dataset(data_path, noise_rate=noise_rate, batch_size = batch_size)
elif args.data == 'aptos':
train_loader, valid_loader = utils.get_aptos_noise_dataset(data_path, noise_rate=noise_rate, batch_size = batch_size)
elif args.data == 'nihchest':
train_loader, valid_loader = utils.get_nihxray(data_path, batch_size = batch_size)
elif args.data == 'idrid':
train_loader, valid_loader = utils.get_idrid_noise_dataset(data_path, noise_rate=noise_rate, batch_size = batch_size)
elif args.data == 'chaoyang':
train_loader, valid_loader = utils.get_chaoyang_dataset(data_path, batch_size = batch_size)
elif 'mnist' in args.data:
train_loader, valid_loader = utils.get_mnist_noise_dataset(args.data, noise_rate=noise_rate, batch_size = batch_size)
elif args.data == 'dr':
train_loader, valid_loader, _ = utils.get_dr(data_path, batch_size = batch_size)
elif 'cifar' in args.data:
train_loader, valid_loader = utils.get_cifar_noise_dataset(args.data, data_path, batch_size = batch_size, noise_rate=noise_rate)
elif args.data == 'clothing':
train_loader, valid_loader = utils.get_clothing1m_dataset(data_path, batch_size=batch_size)
lr_decay = [5, 10]
num_gradual = 1
elif args.data == 'webvision':
train_loader, valid_loader = utils.get_webvision(data_path, batch_size=batch_size)
if args.netsize == 's':
model_load = dino_variant._small_dino
variant = dino_variant._small_variant
elif args.netsize == 'b':
model_load = dino_variant._base_dino
variant = dino_variant._base_variant
elif args.netsize == 'l':
model_load = dino_variant._large_dino
variant = dino_variant._large_variant
dino = torch.hub.load('facebookresearch/dinov2', model_load)
dino_state_dict = dino.state_dict()
model1 = rein.LoRADinoVisionTransformer(dino)
model1.load_state_dict(dino_state_dict, strict=False)
model1.linear = nn.Linear(variant['embed_dim'], config['num_classes'])
model1.to(device)
dino = torch.hub.load('facebookresearch/dinov2', model_load)
dino_state_dict = dino.state_dict()
model2 = rein.LoRADinoVisionTransformer(dino)
model2.load_state_dict(dino_state_dict, strict=False)
model2.linear = nn.Linear(variant['embed_dim'], config['num_classes'])
model2.to(device)
# optimizer = torch.optim.SGD(model.parameters(), lr = 0.01, momentum=0.9, weight_decay = 1e-05)
optimizer = torch.optim.Adam(list(model1.parameters())+ list(model2.parameters()), lr=1e-3)
criterion = others.jocor.loss_jocor
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, lr_decay)
saver = timm.utils.CheckpointSaver(model1, optimizer, checkpoint_dir= save_path, max_history = 1)
print(train_loader.dataset[0][0].shape)
if args.data == 'dr':
num_samples = {0: 25810, 1: 2443, 2: 5292, 3: 873, 4: 708}
class_weight = torch.tensor([1-num_samples[x]/sum(num_samples.values()) for x in num_samples]).to(device)
print(class_weight)
else:
class_weight = None
exponent = 1 # 0.5, 1 or 2; This parameter is equal to c in Tc for R(T) in Co-teaching paper.
rate_schedule = np.ones(max_epoch) * noise_rate
rate_schedule[:num_gradual] = np.linspace(0, noise_rate ** exponent, num_gradual)
# print(ra)
avg_accuracy = 0.0
avg_kappa = 0.0
for epoch in range(max_epoch):
## training
model1.train()
model2.train()
total_loss = 0
total = 0
correct = 0
start_time = time.time()
for batch_idx, (inputs, targets) in enumerate(train_loader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
# Forward + Backward + Optimize
features1 = model1.forward_features(inputs)
outputs1 = model1.linear(features1)
features2 = model2.forward_features(inputs)
outputs2 = model2.linear(features2)
loss_1, loss_2 = criterion(outputs1, outputs2, targets, rate_schedule[epoch])
loss_1.backward()
optimizer.step()
total_loss += loss_1
total += targets.size(0)
_, predicted = outputs1[:len(targets)].max(1)
correct += predicted.eq(targets).sum().item()
print('\r', batch_idx, len(train_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (total_loss/(batch_idx+1), 100.*correct/total, correct, total), end = '')
train_accuracy = correct/total
end_time = time.time()
train_avg_loss = total_loss/len(train_loader)
print()
print(end_time - start_time)
## validation
model1.eval()
model2.eval()
total_loss = 0
total = 0
correct = 0
valid_accuracy = utils.validation_accuracy_lora(model1, valid_loader, device)
if epoch >= max_epoch-10:
avg_accuracy += valid_accuracy
kappa = 1# utils.validation_kohen_kappa(model1, valid_loader, device)
avg_kappa += kappa
scheduler.step()
saver.save_checkpoint(epoch, metric = valid_accuracy)
print('EPOCH {:4}, TRAIN [loss - {:.4f}, acc - {:.4f}], VALID [acc - {:.4f}]\n'.format(epoch, train_avg_loss, train_accuracy, valid_accuracy))
print(scheduler.get_last_lr())
with open(os.path.join(save_path, 'avgacc.txt'), 'w') as f:
f.write(str(avg_accuracy/10))
f.write('|')
f.write(str(avg_kappa/10))
if __name__ =='__main__':
train()