Project has been extended to variational graph autoencoders in salfaris/PriorVGAE.
Variational autoencoder implementation on categorical and continuous datasets using dm-haiku.
The classification example is copied over from the haiku library. The aim is to classify hand-drawn digits to the correct number (a classic).
The continuous version is used for encoding Gaussian process priors, but can be extended to any continuous function.