from copy import deepcopy
if method == 'BayesWRM' or method=='Bayes':
for net in model_list:
for p in net.parameters():
p.requires_grad = False
net.eval()
else:
for p in net.parameters():
p.requires_grad = False
net.eval()
for epsilon in epsilon_set:
train_dataset, val_dataset = utils.get_dataset(params)
loader_test = torch.utils.data.DataLoader(val_dataset,
batch_size=256, shuffle=False)
#test(net, loader_test)
if method == 'BayesWRM' or method == 'Bayes':
adversary = FGSMAttack(model_list, epsilon, is_train=False)
advacc = attack_over_test_data(model_list, adversary, param, loader_test)
else:
adversary = FGSMAttack(net, epsilon, is_train=False, advtraining=method)
advacc = attack_over_test_data(net, adversary, param, loader_test)
print('method',method, 'adv accuracy', advacc)
advacc_set.append(advacc)
epsilon_set = np.linspace(0, 0.3, 11)
advacc_set = []
from copy import deepcopy
if method == 'BayesWRM' or method == 'Bayes':
for net in model_list:
for p in net.parameters():
p.requires_grad = False
net.eval()
else:
for p in net.parameters():
p.requires_grad = False
net.eval()
train_dataset, val_dataset = utils.get_dataset(params,
fixedindex=fixedindex)
loader_test = torch.utils.data.DataLoader(val_dataset,
batch_size=1,
shuffle=False)
#test(net, loader_test)
if method == 'BayesWRM' or method == 'Bayes':
adversary = FGSMAttack(model_list,
epsilon,
is_train=False,
advtraining=method,
inputdist=inputdist,
noiseratio=noiseratio,
storeadv=True)
#adversary = CWAttack(model_list, steps=k)
