def test_gaussian_noise_disable_training_state(self):
test_input = np.ones((50, 20))
gauss_noise = layers.GaussianNoise(std=1)
with gauss_noise.disable_training_state():
layer_output = gauss_noise.output(test_input)
np.testing.assert_array_equal(layer_output, test_input)
def test_gaussian_noise_repr(self):
layer = layers.GaussianNoise(0, 1)
self.assertEqual("GaussianNoise(mean=0, std=1)", str(layer))
def test_gaussian_noise_layer(self):
test_input = np.zeros((50, 20))
gauss_noise = layers.GaussianNoise(std=0.5)
layer_output = self.eval(gauss_noise.output(test_input))
self.assertTrue(stats.mstats.normaltest(layer_output))
for (left_ax, right_ax), real_image, predicted_image in iterator:
real_image = real_image.reshape((28, 28))
predicted_image = predicted_image.reshape((28, 28))
left_ax.imshow(real_image, cmap=plt.cm.binary)
right_ax.imshow(predicted_image, cmap=plt.cm.binary)
plt.show()
if __name__ == '__main__':
autoencoder = algorithms.Momentum(
[
layers.Input(784),
layers.GaussianNoise(mean=0.5, std=0.1),
layers.Sigmoid(100),
layers.Sigmoid(784),
],
step=0.1,
verbose=True,
momentum=0.9,
nesterov=True,
loss='rmse',
)
print("Preparing data...")
x_train, x_test = load_data()
print("Training autoencoder...")
autoencoder.train(x_train, x_train, x_test, x_test, epochs=40)
