-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
143 lines (106 loc) · 6.05 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
import torch
import torch.nn as nn
import torch.optim as optim
# import models.resnet_cifar
import os
import sys
import argparse
import copy
import json
import datetime
import logging
import numpy as np
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
import random
torch.manual_seed(0)
random.seed(0)
np.random.seed(0)
# from art.estimators.classification import PyTorchClassifier
from ModifiedPyTorchClassifier import PyTorchClassifier
# from art.utils import load_dataset
import utils
import pruning_utils
import models.resnet
start_time = datetime.datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%S.%f%Z")
start_time_int = datetime.datetime.now()
args = utils.parse_args()
logger = utils.set_logger(args.output_folder_dir, args)
setting = utils.register_args_to_setting(args, logger)
if args.gpu != '':
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
device = 'cuda' if torch.cuda.is_available() else 'cpu'
logger.info(f"Experiment {setting['management']['exp_desc']} (task: {setting['management']['task']}) started at {start_time}.")
# train_data_gen, x_test, y_test, clip_values, input_shape, nb_classes, trainloader = utils.load_dataset(setting)
train_data_gen, testloader, clip_values, input_shape, nb_classes, trainloader = utils.load_dataset(setting)
logger.info(f"Dataset {setting['management']['dataset']} loaded.")
criterion = nn.CrossEntropyLoss()
# model = models.resnet.resnet20()
if setting['management']['task'] == 'iterative_prune':
model = pruning_utils.load_model(setting["prune_params"]["prune_method_name"],utils.load_model(setting, logger), setting, logger)
vanilla_model = model.model
optimizer = pruning_utils.load_optimizer(setting["prune_params"]["prune_method_name"], model, setting)
else:
model = utils.load_model(setting, logger)
vanilla_model = model
optimizer = optim.SGD(vanilla_model.parameters(), lr = setting['train_params']['lr'], momentum = setting['train_params']['momentum'], weight_decay = setting['train_params']['weight_decay'])
classifier = PyTorchClassifier(
model = vanilla_model,
clip_values = clip_values,
loss = criterion,
optimizer = optimizer,
input_shape = input_shape,
nb_classes = nb_classes,
setting = setting,
)
logger.info(f"Model {setting['management']['model_dir']} successfully wrapped as an ART classifier.")
setting['results'] = dict()
setting['results']['best'] = dict()
if setting['management']['task'] == 'test':
logger.info(f"Testing performance of assigned model {setting['management']['model_dir']}...")
utils.eval_classifier(classifier, epoch_num = 0, testloader = testloader, setting = setting, logger = logger, attack_flag = args.adv_attack)
if setting['management']['task'] == 'train' or setting['management']['task'] == 'finetune':
logger.info(f"Testing baseline performance of model {setting['management']['model_dir']}...")
utils.eval_classifier(classifier, epoch_num = 0, testloader = testloader, setting = setting, logger = logger, attack_flag = args.adv_attack)
logger.info(f"Training classifier for {setting['train_params']['epoch_num']} epochs...")
classifier.fit_generator(train_data_gen, nb_epochs = setting['train_params']['epoch_num'], testloader = testloader, setting = setting, logger = logger, attack_flag = args.adv_attack, trainloader = trainloader)
if setting['management']['task'] == 'train':
utils.save_trained_model(model, 'trained', setting, logger)
elif setting['management']['task'] == 'fintune':
utils.save_trained_model(model, 'fintuned', setting, logger)
else:
logger.info(f"Model {setting['management']['model_dir']} finetuned, please check {setting['management']['save_subdir']}/best_ckpts/ folder for best models.")
utils.pprint_best_models(setting, logger)
elif setting['management']['task'] == 'iterative_prune':
logger.info(f"Training classifier using {setting['prune_params']['prune_method_name']} for {setting['train_params']['epoch_num']} epochs...")
#classifier.fit_generator(train_data_gen, nb_epochs = setting['train_params']['epoch_num'], = x_test, y_test = y_test, setting = setting, logger = logger, attack_flag = args.adv_attack, trainloader = trainloader, shell = model)
classifier.fit_generator(train_data_gen, nb_epochs = setting['train_params']['epoch_num'], testloader = testloader, setting = setting, logger = logger, attack_flag = args.adv_attack, trainloader = trainloader, shell = model)
utils.save_trained_model(classifier.model, 'iterative_pruned', setting, logger)
utils.pprint_best_models(setting, logger)
elif setting['management']['task'] == 'offline_loop_test':
del classifier
eval_worthy_state_dict_full_path_list = utils.get_eval_worthy_epochs(setting, logger)
for eval_epoch_i, eval_epoch_path in eval_worthy_state_dict_full_path_list:
vanilla_model.load_state_dict(torch.load(eval_epoch_path))
classifier = PyTorchClassifier(
model = vanilla_model,
clip_values = clip_values,
loss = criterion,
optimizer = optimizer,
input_shape = input_shape,
nb_classes = nb_classes,
setting = setting,
)
logger.info(f"Epoch #{eval_epoch_i} \t ({eval_epoch_path}) successfully wrapped as an ART classifier.")
utils.eval_classifier(classifier, epoch_num = eval_epoch_i, testloader = testloader, setting = setting, logger = logger, attack_flag = args.adv_attack)
del classifier
logger.info(f"Offline loop tests finished.")
utils.pprint_best_models(setting, logger)
end_time = datetime.datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%S.%f%Z")
end_time_int = datetime.datetime.now()
setting_output_path = setting['management']['output_folder_dir'] + 'output_setting.json'
with open(setting_output_path, "w+") as out_setting_f:
logger.info(f"Saving setting file to {setting_output_path}...")
json.dump(setting, out_setting_f, indent = 4)
total_time = str(end_time_int - start_time_int)
logger.info(f"Experiment {setting['management']['exp_desc']} (task: {setting['management']['task']}) ended at {end_time} duration: {total_time}.")