Building the verification model

Now we have almost all we need to initiate the training session of our shallow autoencoder. However, we are missing one crucial component. While this part is not, strictly speaking, required to train our autoencoder, we must implement it so that we can visually verify whether our autoencoder has truly learned salient features from the training data or not. To do this, we will actually define two additional networks. Don't worry – these two networks are essentially mirror images of the encoder and decoder functions that are present in the autoencoder network we just defined. Hence, all we will be doing is creating a separate encoder and decoder network, which will match the hyperparameters of the encoder and decoder functions from our autoencoder. These two separate networks will be used for prediction only after our autoencoder has been...