-
Notifications
You must be signed in to change notification settings - Fork 3
/
general_main.py
218 lines (196 loc) · 13.2 KB
/
general_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
213
214
215
216
217
218
import os
import sys
import argparse
import random
import numpy as np
import torch
from experiment.run import multiple_run
from utils.utils import boolean_string
from utils.logging import Logger
def main(args):
print(args)
# set up seed
np.random.seed(args.seed)
random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
args.trick = {'labels_trick': args.labels_trick, 'separated_softmax': args.separated_softmax,
'kd_trick': args.kd_trick, 'kd_trick_star': args.kd_trick_star, 'review_trick': args.review_trick,
'ncm_trick': args.ncm_trick}
sys.stdout = Logger(os.path.join(args.logs_dir, args.data+'_'+args.tag, 'log.txt'))
multiple_run(args, store=args.store, save_path=args.save_path)
if __name__ == "__main__":
# Commandline arguments
parser = argparse.ArgumentParser(description="Online Continual Learning PyTorch")
########################General#########################
parser.add_argument('--num_runs', dest='num_runs', default=1, type=int,
help='Number of runs (default: %(default)s)')
parser.add_argument('--seed', dest='seed', default=0, type=int,
help='Random seed')
parser.add_argument('--tag', default='test', type=str)
parser.add_argument('--logs_dir', default='./logs', type=str)
########################Misc#########################
parser.add_argument('--val_size', dest='val_size', default=0.1, type=float,
help='val_size (default: %(default)s)')
parser.add_argument('--num_val', dest='num_val', default=3, type=int,
help='Number of batches used for validation (default: %(default)s)')
parser.add_argument('--num_runs_val', dest='num_runs_val', default=3, type=int,
help='Number of runs for validation (default: %(default)s)')
parser.add_argument('--num_classes_per_task', default=10, type=int)
parser.add_argument('--images_per_class', default=10, type=int)
parser.add_argument('--error_analysis', dest='error_analysis', default=False, type=boolean_string,
help='Perform error analysis (default: %(default)s)')
parser.add_argument('--verbose', type=boolean_string, default=True,
help='print information or not (default: %(default)s)')
parser.add_argument('--store', type=boolean_string, default=False,
help='Store result or not (default: %(default)s)')
parser.add_argument('--save-path', dest='save_path', default=None)
########################Agent#########################
parser.add_argument('--agent', dest='agent', default='ER',
choices=['ER', 'EWC', 'AGEM', 'CNDPM', 'LWF', 'ICARL', 'GDUMB', 'ASER', 'SCR', 'SSCR'],
help='Agent selection (default: %(default)s)')
parser.add_argument('--update', dest='update', default='random', choices=['random', 'GSS', 'ASER', 'summarize'],
help='Update method (default: %(default)s)')
parser.add_argument('--retrieve', dest='retrieve', default='random', choices=['MIR', 'random', 'ASER', 'match', 'mem_match'],
help='Retrieve method (default: %(default)s)')
########################Optimizer#########################
parser.add_argument('--optimizer', dest='optimizer', default='SGD', choices=['SGD', 'Adam'],
help='Optimizer (default: %(default)s)')
parser.add_argument('--learning_rate', dest='learning_rate', default=0.1,
type=float,
help='Learning_rate (default: %(default)s)')
parser.add_argument('--epoch', dest='epoch', default=1,
type=int,
help='The number of epochs used for one task. (default: %(default)s)')
parser.add_argument('--batch', dest='batch', default=10,
type=int,
help='Batch size (default: %(default)s)')
parser.add_argument('--test_batch', dest='test_batch', default=128,
type=int,
help='Test batch size (default: %(default)s)')
parser.add_argument('--weight_decay', dest='weight_decay', type=float, default=0,
help='weight_decay')
########################Data#########################
parser.add_argument('--num_tasks', dest='num_tasks', default=10,
type=int,
help='Number of tasks (default: %(default)s), OpenLORIS num_tasks is predefined')
parser.add_argument('--fix_order', dest='fix_order', default=False,
type=boolean_string,
help='In NC scenario, should the class order be fixed (default: %(default)s)')
parser.add_argument('--plot_sample', dest='plot_sample', default=False,
type=boolean_string,
help='In NI scenario, should sample images be plotted (default: %(default)s)')
parser.add_argument('--data', dest='data', default="cifar10",
help='Path to the dataset. (default: %(default)s)')
parser.add_argument('--cl_type', dest='cl_type', default="nc", choices=['nc', 'ni'],
help='Continual learning type: new class "nc" or new instance "ni". (default: %(default)s)')
parser.add_argument('--ns_factor', dest='ns_factor', nargs='+',
default=(0.0, 0.4, 0.8, 1.2, 1.6, 2.0, 2.4, 2.8, 3.2, 3.6), type=float,
help='Change factor for non-stationary data(default: %(default)s)')
parser.add_argument('--ns_type', dest='ns_type', default='noise', type=str, choices=['noise', 'occlusion', 'blur'],
help='Type of non-stationary (default: %(default)s)')
parser.add_argument('--ns_task', dest='ns_task', nargs='+', default=(1, 1, 2, 2, 2, 2), type=int,
help='NI Non Stationary task composition (default: %(default)s)')
parser.add_argument('--online', dest='online', default=True,
type=boolean_string,
help='If False, offline training will be performed (default: %(default)s)')
########################SSD########################
parser.add_argument('--queue_size', default=64, type=int,
help='Number of images in the queue for information summarizing')
parser.add_argument('--summarize_interval', default=6, type=int,
help='Interval between iterations that summarizing is conducted')
parser.add_argument('--match', default='grad', type=str, choices=['grad', 'feat', 'grad_feat'],
help='The matching target between real and synthesized images')
parser.add_argument('--metric', default='mse', type=str,
help='The distance metric')
parser.add_argument('--lr_img', default=2e-4, type=float,
help='The learning rate for updating images that are being summarized')
parser.add_argument('--mem_weight', default=1, type=float,
help='The weight for relationship matching')
parser.add_argument('--mem_sim_num', default=10, type=int,
help='Number of samples for calculating relationship distance')
parser.add_argument('--mem_extra', default=1, type=int,
help='Whether the final loss is direct sum or weighted sum')
########################ER#########################
parser.add_argument('--mem_size', dest='mem_size', default=10000,
type=int,
help='Memory buffer size (default: %(default)s)')
parser.add_argument('--eps_mem_batch', dest='eps_mem_batch', default=10,
type=int,
help='Episode memory per batch (default: %(default)s)')
########################EWC##########################
parser.add_argument('--lambda', dest='lambda_', default=100, type=float,
help='EWC regularization coefficient')
parser.add_argument('--alpha', dest='alpha', default=0.9, type=float,
help='EWC++ exponential moving average decay for Fisher calculation at each step')
parser.add_argument('--fisher_update_after', dest='fisher_update_after', type=int, default=50,
help="Number of training iterations after which the Fisher will be updated.")
########################MIR#########################
parser.add_argument('--subsample', dest='subsample', default=50,
type=int,
help='Number of subsample to perform MIR(default: %(default)s)')
########################GSS#########################
parser.add_argument('--gss_mem_strength', dest='gss_mem_strength', default=10, type=int,
help='Number of batches randomly sampled from memory to estimate score')
parser.add_argument('--gss_batch_size', dest='gss_batch_size', default=10, type=int,
help='Random sampling batch size to estimate score')
########################ASER########################
parser.add_argument('--k', dest='k', default=5,
type=int,
help='Number of nearest neighbors (K) to perform ASER (default: %(default)s)')
parser.add_argument('--aser_type', dest='aser_type', default="asvm", type=str, choices=['neg_sv', 'asv', 'asvm'],
help='Type of ASER: '
'"neg_sv" - Use negative SV only,'
' "asv" - Use extremal values of Adversarial SV and Cooperative SV,'
' "asvm" - Use mean values of Adversarial SV and Cooperative SV')
parser.add_argument('--n_smp_cls', dest='n_smp_cls', default=2.0,
type=float,
help='Maximum number of samples per class for random sampling (default: %(default)s)')
########################CNDPM#########################
parser.add_argument('--stm_capacity', dest='stm_capacity', default=1000, type=int, help='Short term memory size')
parser.add_argument('--classifier_chill', dest='classifier_chill', default=0.01, type=float,
help='NDPM classifier_chill')
parser.add_argument('--log_alpha', dest='log_alpha', default=-300, type=float, help='Prior log alpha')
########################GDumb#########################
parser.add_argument('--minlr', dest='minlr', default=0.0005, type=float, help='Minimal learning rate')
parser.add_argument('--clip', dest='clip', default=10., type=float,
help='value for gradient clipping')
parser.add_argument('--mem_epoch', dest='mem_epoch', default=70, type=int, help='Epochs to train for memory')
#######################Tricks#########################
parser.add_argument('--labels_trick', dest='labels_trick', default=False, type=boolean_string,
help='Labels trick')
parser.add_argument('--separated_softmax', dest='separated_softmax', default=False, type=boolean_string,
help='separated softmax')
parser.add_argument('--kd_trick', dest='kd_trick', default=False, type=boolean_string,
help='Knowledge distillation with cross entropy trick')
parser.add_argument('--kd_trick_star', dest='kd_trick_star', default=False, type=boolean_string,
help='Improved knowledge distillation trick')
parser.add_argument('--review_trick', dest='review_trick', default=False, type=boolean_string,
help='Review trick')
parser.add_argument('--ncm_trick', dest='ncm_trick', default=False, type=boolean_string,
help='Use nearest class mean classifier')
parser.add_argument('--mem_iters', dest='mem_iters', default=1, type=int,
help='mem_iters')
####################Early Stopping######################
parser.add_argument('--min_delta', dest='min_delta', default=0., type=float,
help='A minimum increase in the score to qualify as an improvement')
parser.add_argument('--patience', dest='patience', default=0, type=int,
help='Number of events to wait if no improvement and then stop the training.')
parser.add_argument('--cumulative_delta', dest='cumulative_delta', default=False, type=boolean_string,
help='If True, `min_delta` defines an increase since the last `patience` reset, '
'otherwise, it defines an increase after the last event.')
####################SupContrast######################
parser.add_argument('--temp', type=float, default=0.07,
help='temperature for loss function')
parser.add_argument('--buffer_tracker', type=boolean_string, default=False,
help='Keep track of buffer with a dictionary')
parser.add_argument('--warmup', type=int, default=4,
help='warmup of buffer before retrieve')
parser.add_argument('--head', type=str, default='mlp',
help='projection head')
args = parser.parse_args()
args.cuda = torch.cuda.is_available()
main(args)