def test_attack_convergence(bn_model, bn_criterion, bn_images, bn_labels):
attack = BoundaryAttack(bn_model, bn_criterion)
advs = attack(bn_images, bn_labels, unpack=False, verbose=True)
for adv in advs:
assert adv.perturbed is not None
assert adv.distance.value < np.inf
attack2 = BoundaryAttack(bn_model, bn_criterion)
bn_images2 = np.array([adv.perturbed for adv in advs])
advs2 = attack2(bn_images2, bn_labels, unpack=False, iterations=5000)
for adv in advs2:
# should converge
assert adv.perturbed is not None
assert adv.distance.value < np.inf
def test_attack(bn_model, bn_criterion, bn_images, bn_labels):
attack = BoundaryAttack(bn_model, bn_criterion)
advs = attack(bn_images,
bn_labels,
unpack=False,
iterations=200,
verbose=True)
for adv in advs:
assert adv.perturbed is not None
assert adv.distance.value < np.inf
def test_attack_parameters3(bn_model, bn_criterion, bn_images, bn_labels):
attack = BoundaryAttack(bn_model, bn_criterion)
np.random.seed(2)
starting_point = np.random.uniform(0, 1, size=bn_images[0].shape).astype(
bn_images.dtype)
advs = attack(bn_images,
bn_labels,
unpack=False,
iterations=200,
starting_point=starting_point,
log_every_n_steps=2,
tune_batch_size=30,
threaded_rnd=False,
threaded_gen=False,
verbose=True)
for adv in advs:
assert adv.perturbed is not None
assert adv.distance.value < np.inf