def test_class_gradient(self):
# Get MNIST
(_, _), (x_test, y_test), _, _ = load_mnist()
x_test = x_test[:NB_TEST]
x_test = np.swapaxes(x_test, 1, 3)
# Test gradient
ptc = PyTorchClassifier(None, self._model, self._loss_fn,
self._optimizer, (1, 28, 28), (10, ))
grads = ptc.class_gradient(x_test)
self.assertTrue(
np.array(grads.shape == (NB_TEST, 10, 1, 28, 28)).all())
self.assertTrue(np.sum(grads) != 0)
def test_class_gradient_target(self):
# Get MNIST
(_, _), (x_test, y_test), _, _ = load_mnist()
x_test = x_test[:NB_TEST]
x_test = np.swapaxes(x_test, 1, 3)
# Create model
model, loss_fn, optimizer = self._model_setup_module()
# Test gradient
ptc = PyTorchClassifier((0, 1), model, loss_fn, optimizer, (1, 28, 28),
10)
grads = ptc.class_gradient(x_test, label=3)
self.assertTrue(np.array(grads.shape == (NB_TEST, 1, 1, 28, 28)).all())
self.assertTrue(np.sum(grads) != 0)
