Implement Categorical Variational autoencoder using Pytorch
Reproducing the results of https://arxiv.org/pdf/1611.01144.pdf (or https://arxiv.org/abs/1611.00712) in Pytorch framework. The original implementation by authors use Tensorflow. In Pytorch there is no readily available Gumbel Softmax distribution to sample from, so have to implement the Relaxed Categorical representation to sample the latent representation.