-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
212 lines (192 loc) · 9.97 KB
/
main.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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
import os
import argparse
import random
import importlib
import numpy as np
import torch
from torch import nn
from torch import optim
from Train import train, train_st
from Test import test
from utils import global_variable as GV
from models import linear_classifier
def display_args(args):
print('===== task arguments =====')
print('data_name = %s' % (args.data_name))
print('network_name = %s' % (args.network_name))
print('model_name = %s' % (args.model_name))
print('N = %d' % (args.N))
print('K = %d' % (args.K))
print('Q = %d' % (args.Q))
print('===== experiment environment arguments =====')
print('devices = %s' % str(args.devices))
print('flag_debug = %r' % (args.flag_debug))
print('n_workers = %d' % (args.n_workers))
print('===== optimizer arguments =====')
print('lr_network = %f' % (args.lr_network))
print('lr = %f' % (args.lr))
print('point = %s' % str(args.point))
print('gamma = %f' % (args.gamma))
print('wd = %f' % (args.wd))
print('mo = %f' % (args.mo))
print('===== training procedure arguments =====')
print('n_training_episodes = %d' % (args.n_training_episodes))
print('n_validating_episodes = %d' % (args.n_validating_episodes))
print('n_testing_episodes = %d' % (args.n_testing_episodes))
print('flag_random_task = %r' % (args.flag_random_task))
print('episode_gap = %d' % (args.episode_gap))
print('===== model arguments =====')
print('tau = %f' % (args.tau))
print('NN = %d' % (args.NN))
print('lambd = %f' % (args.lambd))
if __name__ == '__main__':
# create a parser
parser = argparse.ArgumentParser()
# task arguments
parser.add_argument('--data_name', type=str, default='mini_imagenet', choices=['mini_imagenet', 'tiered_imagenet'])
parser.add_argument('--network_name', type=str, default='resnet', choices=['resnet', 'mlp'])
parser.add_argument('--model_name', type=str, default='protonet', choices=['protonet', 'st'])
parser.add_argument('--N', type=int, default=5)
parser.add_argument('--K', type=int, default=1)
parser.add_argument('--Q', type=int, default=15)
# experiment environment arguments
parser.add_argument('--devices', type=int, nargs='+', default=GV.DEVICES)
parser.add_argument('--flag_debug', action='store_true', default=False)
parser.add_argument('--n_workers', type=int, default=GV.WORKERS)
# optimizer arguments
parser.add_argument('--lr_network', type=float, default=0.001)
parser.add_argument('--lr', type=float, default=0.01)
parser.add_argument('--point', type=int, nargs='+', default=(20,30,40))
parser.add_argument('--gamma', type=float, default=0.2)
parser.add_argument('--wd', type=float, default=0.0005) # weight decay
parser.add_argument('--mo', type=float, default=0.9) # momentum
# training procedure arguments
parser.add_argument('--n_training_episodes', type=int, default=10000) # task list: 20000
parser.add_argument('--n_validating_episodes', type=int, default=1000) # task list: 20000
parser.add_argument('--n_testing_episodes', type=int, default=1000) # task list: 10000
parser.add_argument('--flag_random_task', action='store_true', default=False)
parser.add_argument('--episode_gap', type=int, default=200)
# model arguments
parser.add_argument('--tau', type=float, default=1.0)
parser.add_argument('--NN', type=int, default=64) # number of selected classes
parser.add_argument('--lambd', type=float, default=10) # weight of distill loss in st
args = parser.parse_args()
display_args(args)
data_path = 'datasets/' + args.data_name + '/'
if not args.flag_random_task:
train_task_list_file_path = data_path + 'task_list/' + \
str(args.N) + '-way-' + str(args.K) + '-shot-' + str(args.Q) + '-query-' + 'train.tasklist'
val_task_list_file_path = data_path + 'task_list/' + \
str(args.N) + '-way-' + str(args.K) + '-shot-' + str(args.Q) + '-query-' + 'val.tasklist'
test_task_list_file_path = data_path + 'task_list/' + \
str(args.N) + '-way-' + str(args.K) + '-shot-' + str(args.Q) + '-query-' + 'test.tasklist'
else:
train_task_list_file_path = val_task_list_file_path = test_task_list_file_path = ''
# import modules
Data = importlib.import_module('dataloaders.' + args.data_name)
Network = importlib.import_module('networks.' + args.network_name)
Model = importlib.import_module('models.' + args.model_name)
# generate data loaders
train_data_loader = Data.generate_data_loader(data_path, 'train', args.n_training_episodes,
args.N, args.K + args.Q, args.flag_random_task, train_task_list_file_path)
validate_data_loader = Data.generate_data_loader(data_path, 'validate', args.n_validating_episodes,
args.N, args.K + args.Q, True, val_task_list_file_path)
test_data_loader = Data.generate_data_loader(data_path, 'test', args.n_testing_episodes,
args.N, args.K + args.Q, True, test_task_list_file_path)
print('===== data loader ready. =====')
# generate network
network = Network.MyNetwork(args)
if len(args.devices) > 1:
network = torch.nn.DataParallel(network, device_ids=args.devices)
print('===== network ready. =====')
if args.model_name == 'st':
target_model_save_path = 'saves/finetuned_models/' + args.data_name + '/' + \
'target_model' + '_NN=' + str(args.NN) + '.model'
target_model = torch.load(target_model_save_path)
target_task = target_model['task']
target_params = target_model['state_dict']
target_args = target_model['args']
# generate target network
target_network = Network.MyNetwork(target_args)
if len(args.devices) > 1:
target_network = torch.nn.DataParallel(target_network, device_ids=args.devices)
print('===== target network ready. =====')
# generate target model
target_model = linear_classifier.MyModel(target_args, target_network, target_args.NN)
target_model.load_state_dict(target_params)
target_model = target_model.cuda(args.devices[0])
print('===== target model ready. =====')
# generate model
if args.model_name == 'st':
model = Model.MyModel(args, network, target_model, target_task)
else:
model = Model.MyModel(args, network)
if args.data_name != 'gaussian':
pretrained_file_path = 'saves/pretrained_weights/' + args.data_name + '/' + args.network_name + '.pth'
pretrained_state_dict = torch.load(pretrained_file_path)['state_dict']
pretrained_state_dict = {k:v for k, v in pretrained_state_dict.items() if k.startswith('encoder')}
model_state_dict = model.state_dict()
model_state_dict.update(pretrained_state_dict)
model.load_state_dict(model_state_dict)
model = model.cuda(args.devices[0])
print('===== model ready. =====')
model_save_path = 'saves/trained_models/' + args.data_name + '/' + \
args.network_name + '_' + args.model_name + \
'_N=' + str(args.N) + \
'_K=' + str(args.K) + \
'_Q=' + str(args.Q) + \
'_lr-net=' + str(args.lr_network) + \
'_lr=' + str(args.lr) + \
'_point=' + str(args.point) + \
'_gamma=' + str(args.gamma) + \
'_wd=' + str(args.wd) + \
'_mo=' + str(args.mo) + \
'_randtask=' + str(args.flag_random_task) + \
'_tau=' + str(args.tau) + \
'_NN=' + str(args.NN) + \
'_lambd=' + str(args.lambd) + \
'.model'
statistic_save_path = 'saves/statistics/' + args.data_name + '/' + \
args.network_name + '_' + args.model_name + \
'_N=' + str(args.N) + \
'_K=' + str(args.K) + \
'_Q=' + str(args.Q) + \
'_lr-net=' + str(args.lr_network) + \
'_lr=' + str(args.lr) + \
'_point=' + str(args.point) + \
'_gamma=' + str(args.gamma) + \
'_wd=' + str(args.wd) + \
'_mo=' + str(args.mo) + \
'_randtask=' + str(args.flag_random_task) + \
'_tau=' + str(args.tau) + \
'_NN=' + str(args.NN) + \
'_lambd=' + str(args.lambd) + \
'.stat'
# create directories
dirs = os.path.dirname(model_save_path)
os.makedirs(dirs, exist_ok=True)
dirs = os.path.dirname(statistic_save_path)
os.makedirs(dirs, exist_ok=True)
# training process
if args.model_name == 'st':
training_loss_list, validating_accuracy_list = train_st(args, train_data_loader, validate_data_loader,
model, model_save_path, target_task)
else:
training_loss_list, validating_accuracy_list = train(args, train_data_loader, validate_data_loader,
model, model_save_path)
print('===== training finish. =====')
if not args.flag_debug:
record = {
'training_loss': training_loss_list,
'validating_accuracy': validating_accuracy_list
}
torch.save(record, statistic_save_path)
display_args(args)
# load best model
if not args.flag_debug:
record = torch.load(model_save_path)
model.load_state_dict(record['state_dict'])
print('best model loaded, validating acc = %f' % record['validating_accuracy'])
# testing process
testing_accuracy = test(args, test_data_loader, model)
print('testing acc = %f' % (testing_accuracy))