def test_fit_generator(self):
import torch
from torch.utils.data import DataLoader
from art.data_generators import PyTorchDataGenerator
(x_train, y_train), (x_test, y_test) = self.mnist
acc = np.sum(
np.argmax(self.module_classifier.predict(x_test), axis=1) ==
np.argmax(y_test, axis=1)) / NB_TEST
logger.info('Accuracy: %.2f%%', (acc * 100))
# Create tensors from data
x_train_tens = torch.from_numpy(x_train)
x_train_tens = x_train_tens.float()
y_train_tens = torch.from_numpy(y_train)
# Create PyTorch dataset and loader
dataset = torch.utils.data.TensorDataset(x_train_tens, y_train_tens)
data_loader = DataLoader(dataset=dataset, batch_size=5, shuffle=True)
data_gen = PyTorchDataGenerator(data_loader,
size=NB_TRAIN,
batch_size=5)
# Fit model with generator
self.module_classifier.fit_generator(data_gen, nb_epochs=2)
acc2 = np.sum(
np.argmax(self.module_classifier.predict(x_test), axis=1) ==
np.argmax(y_test, axis=1)) / NB_TEST
logger.info('Accuracy: %.2f%%', (acc * 100))
self.assertTrue(acc2 >= .8 * acc)
def _image_data_generator(**kwargs):
(x_train_mnist, y_train_mnist), (_, _) = get_default_mnist_subset
if framework == "keras" or framework == "kerastf":
return KerasDataGenerator(
iterator=image_iterator,
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "tensorflow":
if not is_tf_version_2:
return TensorFlowDataGenerator(
sess=kwargs["sess"],
iterator=image_iterator,
iterator_type="initializable",
iterator_arg={},
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "pytorch":
return PyTorchDataGenerator(iterator=image_iterator,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
if framework == "mxnet":
return MXDataGenerator(iterator=image_iterator,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
def test_fit_generator(self):
classifier = get_image_classifier_pt()
accuracy = (np.sum(
np.argmax(classifier.predict(self.x_test_mnist), axis=1) ==
np.argmax(self.y_test_mnist, axis=1)) / self.n_test)
logger.info("Accuracy: %.2f%%", (accuracy * 100))
# Create tensors from data
x_train_tens = torch.from_numpy(self.x_train_mnist)
x_train_tens = x_train_tens.float()
y_train_tens = torch.from_numpy(self.y_train_mnist)
# Create PyTorch dataset and loader
dataset = torch.utils.data.TensorDataset(x_train_tens, y_train_tens)
data_loader = DataLoader(dataset=dataset, batch_size=5, shuffle=True)
data_gen = PyTorchDataGenerator(data_loader,
size=self.n_train,
batch_size=5)
# Fit model with generator
classifier.fit_generator(data_gen, nb_epochs=2)
accuracy_2 = (np.sum(
np.argmax(classifier.predict(self.x_test_mnist), axis=1) ==
np.argmax(self.y_test_mnist, axis=1)) / self.n_test)
logger.info("Accuracy: %.2f%%", (accuracy_2 * 100))
self.assertEqual(accuracy, 0.32)
self.assertAlmostEqual(accuracy_2, 0.75, delta=0.1)
class TestPyTorchGenerator(unittest.TestCase):
def setUp(self):
import torch
from torch.utils.data import DataLoader
master_seed(seed=42)
class DummyDataset(torch.utils.data.Dataset):
def __init__(self):
self._size = 10
self._x = np.random.rand(self._size, 1, 5, 5)
self._y = np.random.randint(0, high=10, size=self._size)
def __len__(self):
return self._size
def __getitem__(self, idx):
return self._x[idx], self._y[idx]
dataset = DummyDataset()
data_loader = DataLoader(dataset=dataset, batch_size=5, shuffle=True)
self.data_gen = PyTorchDataGenerator(data_loader,
size=10,
batch_size=5)
def test_gen_interface(self):
x, y = self.data_gen.get_batch()
# Check return types
self.assertTrue(isinstance(x, np.ndarray))
self.assertTrue(isinstance(y, np.ndarray))
# Check shapes
self.assertEqual(x.shape, (5, 1, 5, 5))
self.assertEqual(y.shape, (5, ))
def test_pytorch_specific(self):
import torch
iter_ = iter(self.data_gen.iterator)
x, y = next(iter_)
# Check return types
self.assertTrue(isinstance(x, torch.Tensor))
self.assertTrue(isinstance(y, torch.Tensor))
# Check shapes
self.assertEqual(x.shape, (5, 1, 5, 5))
self.assertEqual(y.shape, (5, ))
def test_error(self):
with self.assertRaises(TypeError):
self.data_gen = PyTorchDataGenerator("data_loader",
size=10,
batch_size=5)
def setUp(self):
import torch
from torch.utils.data import DataLoader
master_seed(42)
class DummyDataset(torch.utils.data.Dataset):
def __init__(self):
self._size = 10
self._x = np.random.rand(self._size, 1, 5, 5)
self._y = np.random.randint(0, high=10, size=self._size)
def __len__(self):
return self._size
def __getitem__(self, idx):
return self._x[idx], self._y[idx]
dataset = DummyDataset()
data_loader = DataLoader(dataset=dataset, batch_size=5, shuffle=True)
self.data_gen = PyTorchDataGenerator(data_loader, size=10, batch_size=5)
def _image_data_generator(**kwargs):
(x_train_mnist, y_train_mnist), (_, _) = get_default_mnist_subset
image_it = image_iterator()
data_generator = None
if framework == "keras" or framework == "kerastf":
data_generator = KerasDataGenerator(
iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "tensorflow1":
data_generator = TensorFlowDataGenerator(
sess=kwargs["sess"],
iterator=image_it,
iterator_type="initializable",
iterator_arg={},
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "pytorch":
data_generator = PyTorchDataGenerator(
iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
if framework == "mxnet":
data_generator = MXDataGenerator(iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
if data_generator is None:
raise ARTTestFixtureNotImplemented(
"framework {0} does not current have any data generator implemented",
image_data_generator.__name__,
framework,
)
return data_generator
def _image_data_generator(**kwargs):
(x_train_mnist, y_train_mnist), (_, _) = get_default_mnist_subset
image_it = image_iterator()
data_generator = None
if framework == "keras" or framework == "kerastf":
data_generator = KerasDataGenerator(
iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "tensorflow1":
data_generator = TensorFlowDataGenerator(
sess=kwargs["sess"],
iterator=image_it,
iterator_type="initializable",
iterator_arg={},
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "tensorflow2":
data_generator = TensorFlowV2DataGenerator(
iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size,
)
if framework == "pytorch":
data_generator = PyTorchDataGenerator(
iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
if framework == "mxnet":
data_generator = MXDataGenerator(iterator=image_it,
size=x_train_mnist.shape[0],
batch_size=default_batch_size)
return data_generator
eps=8.0 / 255.0,
eps_step=2.0 / 255.0,
max_iter=40,
targeted=False,
num_random_init=5,
batch_size=32,
)
# Step 4: Create the trainer object - AdversarialTrainerFBFPyTorch
# if you have apex installed, change use_amp to True
epsilon = 8.0 / 255.0
trainer = AdversarialTrainerFBFPyTorch(classifier, eps=epsilon, use_amp=False)
# Build a Keras image augmentation object and wrap it in ART
art_datagen = PyTorchDataGenerator(iterator=dataloader,
size=x_train.shape[0],
batch_size=128)
# Step 5: fit the trainer
trainer.fit_generator(art_datagen, nb_epochs=30)
x_test_pred = np.argmax(classifier.predict(x_test), axis=1)
print(
"Accuracy on benign test samples after adversarial training: %.2f%%" %
(np.sum(x_test_pred == np.argmax(y_test, axis=1)) / x_test.shape[0] * 100))
x_test_attack = attack.generate(x_test)
x_test_attack_pred = np.argmax(classifier.predict(x_test_attack), axis=1)
print(
"Accuracy on original PGD adversarial samples after adversarial training: %.2f%%"
% (np.sum(x_test_attack_pred == np.argmax(y_test, axis=1)) /
def test_error(self):
with self.assertRaises(TypeError):
self.data_gen = PyTorchDataGenerator("data_loader",
size=10,
batch_size=5)
def fit_generator(self, train_generator, epochs):
train_generator_art = PyTorchDataGenerator(
train_generator, len(train_generator.dataset),
train_generator.batch_size)
self.adversarial_trainer.fit_generator(train_generator_art,
nb_epochs=epochs)