Scalable Modeling of Spatiotemporal Data using the Variational Autoencoder: an Application in Glaucoma.