Let's plot the weights visually and plot them using Matplotlib:
wts = autoencoder.get_weights()dim = math.ceil(math.sqrt(autoencoder.num_hidden))plt.figure(1, figsize=(dim, dim))for i in range(0, autoencoder.num_hidden): im = wts.flatten()[i::autoencoder.num_hidden].reshape((28, 28)) ax = plt.subplot(dim, dim, i + 1) for label in (ax.get_xticklabels() + ax.get_yticklabels()): label.set_fontsize(8) #plt.title('Feature Weights ' + str(i)) plt.imshow(im, cmap="gray", clim=(-1.0, 1.0))plt.suptitle('Additive Gaussian Noise AutoEncoder Weights', fontsize=15, y=0.95)plt.savefig('figures/additive_gaussian_weights.png')plt.show()
Weights of the neurons in hidden layers for the Additive Gaussian Auto Encoder
