batch_size_test=args.batch_size_test)
model = ConvNet1(input_size=[28, 28], input_channels=1, output_class=10)
device_ids, model = parse_device_alloc(device_config=None, model=model)
lr_func = parse_lr(policy=args.lr_policy, epoch_num=args.epoch_num)
optimizer = parse_optim(policy=args.optim_policy,
params=model.parameters())
setup_config = {kwarg: value for kwarg, value in args._get_kwargs()}
setup_config['lr_list'] = [lr_func(idx) for idx in range(args.epoch_num)]
if not os.path.exists(args.output_folder):
os.makedirs(args.output_folder)
tricks = {}
if args.snapshots != None:
tricks['snapshots'] = args.snapshots
results = train_test(setup_config=setup_config,
model=model,
train_loader=train_loader,
test_loader=test_loader,
epoch_num=args.epoch_num,
optimizer=optimizer,
lr_func=lr_func,
output_folder=args.output_folder,
model_name=args.model_name,
device_ids=device_ids,
**tricks)
'test_safe_distance': {},
'guaranteed_distances': []
}
for param in list(sorted(tosave['setup_config'].keys())):
print('%s\t=>%s' % (param, tosave['setup_config'][param]))
# Train
train_test(model=model,
train_loader=train_loader,
test_loader=test_loader,
attacker=attacker,
epoch_num=args.epochs,
optimizer=optimizer,
out_folder=args.out_folder,
model_name=args.model_name,
bound_est=args.bound_est,
alpha_list=alpha_list,
eps_list=eps_list,
gamma_list=gamma_list,
T=args.T,
norm=norm,
device=device,
criterion=criterion,
tosave=tosave,
at_per=args.at_per,
pixel_range=args.pixel_range,
update_freq=args.update_freq,
bound_calc_per_batch=args.bound_calc_per_batch,
regularize_mode=args.regularize_mode)