parser.add_argument('-d', '--data-configs', required=False,
default='../configs/data-minerva-mnist.json',
help='Folder where test data stored in.')
parser.add_argument('-o', '--output-root', required=False,
default='../results',
help='Folder for outputs.')
parser.add_argument('--debug', required=False, default=True)
parser.add_argument('-s', '--save-results', required=False, default=True, help='Save output or not')
args = parser.parse_args()
print('------AUGMENT SUMMARY-------')
print('TRANSFORMATION CONFIGS:', args.trans_configs)
print('MODEL CONFIGS:', args.model_configs)
print('DATA CONFIGS:', args.data_configs)
print('OUTPUT ROOT:', args.output_root)
print('DEBUGGING MODE:', args.debug)
print('SAVING OUTPUT:', args.save_results)
print('----------------------------\n')
# parse configurations (into a dictionary) from json file
trans_configs = load_from_json(args.trans_configs)
model_configs = load_from_json(args.model_configs)
data_configs = load_from_json(args.data_configs)
# -------- test transformations -------------
evaluate(trans_configs=trans_configs,
model_configs=model_configs,
data_configs=data_configs,
save=args.save_results,
output_dir=args.output_root)
"""
parser.add_argument('-d',
'--data-configs',
required=False,
default='../configs/demo/data-mnist.json',
help='Folder where test data stored in.')
parser.add_argument('-o',
'--output-root',
required=False,
default='../../task1_data',
help='Folder for outputs.')
parser.add_argument('--debug', required=False, default=True)
args = parser.parse_args()
data_configs = load_from_json(args.data_configs)
# load the benign samples
bs_file = os.path.join(data_configs.get('dir'),
data_configs.get('bs_file'))
x_bs = np.load(bs_file)
img_rows, img_cols = x_bs.shape[1], x_bs.shape[2]
# load the corresponding true labels
label_file = os.path.join(data_configs.get('dir'),
data_configs.get('label_file'))
labels = np.load(label_file)
# get random subsamples
# for MNIST, num_classes is 10
# files "subsamples-mnist-ratio_0.1-xxxxxx.npy" and "sublabels-mnist-ratio_0.1-xxxxxx.npy"
# will be generated and saved at "/results" folder, where "xxxxxx" are timestamps.
print(f'EXPERIMENT ROOT: {args.experiment_root}')
print(f'POOL CONFIGS: {args.pool_configs}')
print(f'SELECTED POOL: {args.selected_pool}')
print(f'MODEL CONFIGS: {args.model_configs}')
print(f'DATA CONFIGS: {args.data_configs}')
print(f'BENIGN SAMPLES: {args.benign_samples}')
print(f'ATTACK CONFIGS: {args.attack_configs}')
print(f'SELECTED ATTACK: {args.selected_attacks}')
print(f'TARGETED MODEL: {args.targeted_model}')
print(f'EOT or not: {args.eot}')
print(f'OUTPUT ROOT: {args.output_root}')
print(f'DEBUGGING MODE: {args.debug}')
print('-------------------------------\n')
# parse configurations from json file
pool_configs = load_from_json(args.pool_configs)
model_configs = load_from_json(args.model_configs)
model_configs['wresnet']['dir'] = args.experiment_root + model_configs.get(
'wresnet').get('dir')
model_configs['shake26']['dir'] = args.experiment_root + model_configs.get(
'wresnet').get('dir')
data_configs = load_from_json(args.data_configs)
data_configs['dir'] = args.experiment_root + data_configs.get('dir')
attack_configs = load_from_json(args.attack_configs)
# load the targeted model
if args.targeted_model == 'single':
prefix = f'AE-cifar100-wresnet-{args.targeted_model}'
model_file = os.path.join(
model_configs.get('wresnet').get('dir'),
model_configs.get('wresnet').get('um_file'))
default='../../experiment/demo/results',
help='Folder for outputs.')
args = parser.parse_args()
print("------AUGMENT SUMMARY-------")
print("EXPERIMENT ROOT:", args.experiment_root)
print("MODEL CONFIGS:", args.model_configs)
print("OUTPUT ROOT:", args.output_root)
print('----------------------------\n')
# ----------------------------
# parse configurations (into a dictionary) from json file
# ----------------------------
model_configs = load_from_json(args.model_configs)
model_configs["wresnet"]["dir"] = args.experiment_root + model_configs.get("wresnet").get("dir")
# ---------------------------
# load the targeted model
# ---------------------------
# In the context of the adversarially trained model,
# we use the undefended model as adversary's target model.
savefile = "AdvTrained-cifar100.pth"
model_file = os.path.join(model_configs.get("wresnet").get('dir'), model_configs.get("wresnet").get("pgd_trained_cifar"))
model, _, _ = load_model(file=model_file, model_configs=model_configs.get("wresnet"), trans_configs=None)
(x_train, y_train), _ = load_data('cifar100')
pgd_adv_train(model=model,
data=(x_train, y_train),