-
Notifications
You must be signed in to change notification settings - Fork 2
/
main_npc.py
169 lines (145 loc) · 7.03 KB
/
main_npc.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
import argparse
import os
import numpy as np
import torch
####################################################################################################################
parser = argparse.ArgumentParser(description="main")
# data
parser.add_argument('--dataset', type=str, default='MNIST', help = 'MNIST, FMNIST, CIFAR10, Clothing, Food')
parser.add_argument('--noise_type', type=str, default='clean', help='clean, sym, asym, idn, idnx')
parser.add_argument('--noisy_ratio', type=float, default=0.4, help='between 0 and 1')
# classifier
parser.add_argument('--class_method', type=str, default=None)
parser.add_argument('--post_method', type=str, default=None)
# prior mode
parser.add_argument('--knn_mode', type=str, default=None, help='onehot, proba')
parser.add_argument('--selected_class', type=str, default='1',help='2,5,10, ...n_class')
# experiment condition for generator
parser.add_argument('--prior_norm', type=float, default=5, help='rho')
parser.add_argument('--beta', type=float, default=1.0, help='coefficient on kl loss, beta vae')
parser.add_argument("--total_iter", type=int, default=10, help='total iter (Default : 10)')
# general experiment condition
parser.add_argument('--seed', type=int, default=0)
parser.add_argument("--lr", type=float, default=0.001, help = "Learning rate (Default : 1e-3)")
parser.add_argument('--softplus_beta', type=float, default=1, help='softplus beta')
parser.add_argument('--clip_gradient_norm', type=float, default=100000, help='max norm for gradient clipping')
# etc
parser.add_argument('--set_gpu', type=int, default=0, help='gpu setting 0/1/2/3')
parser.add_argument('--data_dir', type=str, default=None)
####################################################################################################################
if __name__ == '__main__':
args = parser.parse_args()
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = str(args.set_gpu)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
np.random.seed(args.seed)
# data and network
if args.dataset in ['MNIST','FMNIST','CIFAR10']:
args.n_classes = 10
args.dropout = 0.3
args.causalnl_z_dim = 25
args.batch_size = 128
if args.dataset in ['MNIST', 'FMNIST']:
args.model = 'CNN_MNIST'
elif args.dataset == 'CIFAR10':
args.model = 'CNN_CIFAR'
elif args.dataset in ['Clothing', 'Food']:
if args.dataset == 'Clothing':
args.n_classes = 14
elif args.dataset == 'Food':
args.n_classes = 101
args.noise_type = 'clean'
args.dropout = 0.0
args.causalnl_z_dim = 100
args.batch_size = 32
args.model = 'Resnet50Pre'
else: # wrong dataset
args.n_classes = None
args.dropout = None
args.batch_size = None
# classifier model
if args.dataset in ['MNIST', 'FMNIST', 'CIFAR10']:
args.pre_epoch = 10
if args.class_method in ['cores', 'rel'] :
args.total_epochs = 100
elif args.class_method =='causalNL':
args.total_epochs = 150
args.dropout = 0.0
else:
args.total_epochs = 200
elif args.dataset in ['Food']:
args.pre_epoch = 5
if args.class_method in ['cores', 'rel']:
args.total_epochs = 50
else:
args.total_epochs = 100
else: # Clothing
args.pre_epoch = 1
args.total_epochs = 10
# directory
if args.noise_type == 'clean':
args.data_name = args.dataset + '_00.0_' + args.noise_type + '.pk'
data_noise = args.dataset + '_' + args.model + '_clean'
else:
args.data_name = args.dataset + '_' + str(100 * args.noisy_ratio) + '_' + args.noise_type + '.pk'
data_noise = args.dataset + '_' + args.model + '_' + args.noise_type + '_' + str(100 * args.noisy_ratio)
if args.dataset in ['Clothing', 'Food']:
args.data_name = args.dataset
# directory for loading trained classifier
data_leaf = args.class_method+'_pre_epoch_'+str(args.pre_epoch)+'_epoch_'+str(args.total_epochs)+'_seed_0'
args.cls_dir = os.path.join('classifier_model', 'result', data_noise, data_leaf) + '/'
args.model_dir = os.path.join('classifier_model','result_model',data_noise,args.class_method)\
+'/pre_epoch_'+str(args.pre_epoch)+ '_epoch_'+str(args.total_epochs)+'_dropout_ratio_'+str(args.dropout * 100)+'_seed_0_'
# post model
if args.post_method == 'rog':
from library.rog import ROG
model = ROG(args)
acc = model.run()
cls_dir = os.path.join('classifier_model', 'result', data_noise,
'rog_pre_epoch_' + str(args.pre_epoch) + '_epoch_' + str(args.total_epochs) +
'_seed_' + str(args.seed) + '_pre_method_' + str(args.class_method)) + '/'
os.makedirs(cls_dir, exist_ok=True)
f = open(cls_dir + "_Acc.txt", "w")
f.write('======================================')
f.write('\n')
f.write("test acc : " + str(acc))
f.write('\n')
f.close()
elif args.post_method == 'knn':
from library.knn_test import KNN_tester
func = KNN_tester(args)
func.knn_test()
else:
from train_npc import NPC
from evaluate import Acc_calculator
from library.util.utils import save_gen_text_configuration_v2
if args.knn_mode == 'onehot':
args.selected_class = '1'
# generate directory to save npc result
args.gen_dir = os.path.join('result', data_noise, data_leaf) \
+ '/beta_' + str(args.beta) \
+ '_prior_norm_' + str(args.prior_norm) \
+ '_gen_epoch_' + str(args.total_iter) \
+ '_seed_' + str(args.seed) \
+ '_act_relu_clip_' + str(args.clip_gradient_norm) \
+ '_softp_' + str(args.softplus_beta) \
+ '/' + args.knn_mode + '_' + args.selected_class + '/'
args.gen_model_dir = os.path.join('result_model', data_noise, args.class_method) \
+ '/beta_' + str(args.beta) \
+ '_prior_norm_' + str(args.prior_norm) \
+ '_gen_epoch_' + str(args.total_iter) \
+ '_seed_' + str(args.seed) \
+ '_act_relu_clip_' + str(args.clip_gradient_norm) \
+ '_softp_' + str(args.softplus_beta) \
+ '_knn_' + args.knn_mode + '_' + args.selected_class
# npc directory
os.makedirs(args.gen_dir, exist_ok=True)
os.makedirs(os.path.join('result_model', data_noise, args.class_method), exist_ok=True)
scaler = torch.cuda.amp.GradScaler()
gen_model = NPC(args, scaler)
gen_model.run()
# model performance calculation process
func = Acc_calculator(args)
accuracy = func.merge_classifier_and_autoencoder()
save_gen_text_configuration_v2(args, accuracy)