def test_minimal_perturbations_images(fix_get_mnist_subset,
get_image_classifier_list_for_attack):
classifier_list = get_image_classifier_list_for_attack(FastGradientMethod)
# TODO this if statement must be removed once we have a classifier for both image and tabular data
if classifier_list is None:
logging.warning(
"Couldn't perform this test because no classifier is defined")
return
for classifier in classifier_list:
attack = FastGradientMethod(classifier, eps=1.0, batch_size=11)
attack_params = {"minimal": True, "eps_step": 0.1, "eps": 5.0}
attack.set_params(**attack_params)
expected_values = {
"x_test_mean":
ExpectedValue(0.03896513, 0.01),
"x_test_min":
ExpectedValue(-0.30000000, 0.00001),
"x_test_max":
ExpectedValue(0.30000000, 0.00001),
"y_test_pred_adv_expected":
ExpectedValue(np.asarray([4, 2, 4, 7, 0, 4, 7, 2, 0, 7, 0]), 2),
}
backend_check_adverse_values(attack, fix_get_mnist_subset,
expected_values)
def test_norm_images(art_warning, norm, fix_get_mnist_subset, image_dl_estimator_for_attack):
try:
classifier = image_dl_estimator_for_attack(FastGradientMethod)
if norm == np.inf:
expected_values = {
"x_test_mean": ExpectedValue(0.2346725, 0.002),
"x_test_min": ExpectedValue(-1.0, 0.00001),
"x_test_max": ExpectedValue(1.0, 0.00001),
"y_test_pred_adv_expected": ExpectedValue(np.asarray([4, 4, 4, 7, 7, 4, 7, 2, 2, 3, 0]), 2),
}
elif norm == 1:
expected_values = {
"x_test_mean": ExpectedValue(0.00051374, 0.002),
"x_test_min": ExpectedValue(-0.01486498, 0.001),
"x_test_max": ExpectedValue(0.014761963, 0.001),
"y_test_pred_adv_expected": ExpectedValue(np.asarray([7, 1, 1, 4, 4, 1, 4, 4, 4, 4, 4]), 4),
}
elif norm == 2:
expected_values = {
"x_test_mean": ExpectedValue(0.007636416, 0.001),
"x_test_min": ExpectedValue(-0.211054801, 0.001),
"x_test_max": ExpectedValue(0.209592223, 0.001),
"y_test_pred_adv_expected": ExpectedValue(np.asarray([7, 2, 4, 4, 4, 7, 7, 4, 0, 4, 4]), 2),
}
attack = FastGradientMethod(classifier, eps=1.0, norm=norm, batch_size=128)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
except ARTTestException as e:
art_warning(e)
def test_one_shot(fix_get_mnist_subset, image_dl_estimator_for_attack):
classifier_list = image_dl_estimator_for_attack(FastGradientMethod)
# TODO this if statement must be removed once we have a classifier for both image and tabular data
if classifier_list is None:
logging.warning(
"Couldn't perform this test because no classifier is defined")
return
# for the one-shot method, frame saliency attack should resort to plain FastGradientMethod
expected_values = {
"x_test_mean":
ExpectedValue(0.2346725, 0.002),
"x_test_min":
ExpectedValue(-1.0, 0.00001),
"x_test_max":
ExpectedValue(1.0, 0.00001),
"y_test_pred_adv_expected":
ExpectedValue(np.asarray([4, 4, 4, 7, 7, 4, 7, 2, 2, 3, 0]), 2),
}
for classifier in classifier_list:
attacker = FastGradientMethod(classifier, eps=1, batch_size=128)
attack = FrameSaliencyAttack(classifier, attacker, "one_shot")
backend_check_adverse_values(attack, fix_get_mnist_subset,
expected_values)
def test_minimal_perturbations_images(art_warning, fix_get_mnist_subset, image_dl_estimator_for_attack):
try:
classifier = image_dl_estimator_for_attack(FastGradientMethod)
expected_values = {
"x_test_mean": ExpectedValue(0.03896513, 0.01),
"x_test_min": ExpectedValue(-0.30000000, 0.00001),
"x_test_max": ExpectedValue(0.30000000, 0.00001),
"y_test_pred_adv_expected": ExpectedValue(np.asarray([4, 2, 4, 7, 0, 4, 7, 2, 0, 7, 0]), 2),
}
attack = FastGradientMethod(classifier, eps=1.0, batch_size=11)
# Test eps of float type
attack_params = {"minimal": True, "eps_step": 0.1, "eps": 5.0}
attack.set_params(**attack_params)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
# Test eps of array type 1
(_, _, x_test_mnist, _) = fix_get_mnist_subset
eps = np.ones(shape=x_test_mnist.shape) * 5.0
eps_step = np.ones_like(eps) * 0.1
attack_params = {"minimal": True, "eps_step": eps_step, "eps": eps}
attack.set_params(**attack_params)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
# Test eps of array type 2
eps = np.ones(shape=x_test_mnist.shape[1:]) * 5.0
eps_step = np.ones_like(eps) * 0.1
attack_params = {"minimal": True, "eps_step": eps_step, "eps": eps}
attack.set_params(**attack_params)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
# Test eps of array type 3
eps = np.ones(shape=x_test_mnist.shape[2:]) * 5.0
eps_step = np.ones_like(eps) * 0.1
attack_params = {"minimal": True, "eps_step": eps_step, "eps": eps}
attack.set_params(**attack_params)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
# Test eps of array type 4
eps = np.ones(shape=x_test_mnist.shape[3:]) * 5.0
eps_step = np.ones_like(eps) * 0.1
attack_params = {"minimal": True, "eps_step": eps_step, "eps": eps}
attack.set_params(**attack_params)
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
except ARTTestException as e:
art_warning(e)
def test_norm_images(norm, fix_get_mnist_subset,
get_image_classifier_list_for_attack):
classifier_list = get_image_classifier_list_for_attack(FastGradientMethod)
# TODO this if statement must be removed once we have a classifier for both image and tabular data
if classifier_list is None:
logging.warning(
"Couldn't perform this test because no classifier is defined")
return
elif norm == np.inf:
expected_values = {
"x_test_mean":
ExpectedValue(0.2346725, 0.002),
"x_test_min":
ExpectedValue(-1.0, 0.00001),
"x_test_max":
ExpectedValue(1.0, 0.00001),
"y_test_pred_adv_expected":
ExpectedValue(np.asarray([4, 4, 4, 7, 7, 4, 7, 2, 2, 3, 0]), 2),
}
elif norm == 1:
expected_values = {
"x_test_mean":
ExpectedValue(0.00051374, 0.002),
"x_test_min":
ExpectedValue(-0.01486498, 0.001),
"x_test_max":
ExpectedValue(0.014761963, 0.001),
"y_test_pred_adv_expected":
ExpectedValue(np.asarray([7, 1, 1, 4, 4, 1, 4, 4, 4, 4, 4]), 4),
}
elif norm == 2:
expected_values = {
"x_test_mean":
ExpectedValue(0.007636416, 0.001),
"x_test_min":
ExpectedValue(-0.211054801, 0.001),
"x_test_max":
ExpectedValue(0.209592223, 0.001),
"y_test_pred_adv_expected":
ExpectedValue(np.asarray([7, 2, 4, 4, 4, 7, 7, 4, 0, 4, 4]), 2),
}
for classifier in classifier_list:
attack = FastGradientMethod(classifier,
eps=1,
norm=norm,
batch_size=128)
backend_check_adverse_values(attack, fix_get_mnist_subset,
expected_values)
def test_one_shot(art_warning, fix_get_mnist_subset, image_dl_estimator_for_attack):
try:
classifier = image_dl_estimator_for_attack(FastGradientMethod)
# for the one-shot method, frame saliency attack should resort to plain FastGradientMethod
expected_values = {
"x_test_mean": ExpectedValue(0.2346725, 0.002),
"x_test_min": ExpectedValue(-1.0, 0.00001),
"x_test_max": ExpectedValue(1.0, 0.00001),
"y_test_pred_adv_expected": ExpectedValue(np.asarray([4, 4, 4, 7, 7, 4, 7, 2, 2, 3, 0]), 2),
}
attacker = FastGradientMethod(classifier, eps=1.0, batch_size=128)
attack = FrameSaliencyAttack(classifier, attacker, "one_shot")
backend_check_adverse_values(attack, fix_get_mnist_subset, expected_values)
except ARTTestException as e:
art_warning(e)
