Which order of operations should I use for creating a Convolutional Autoencoder in Keras