Dustin Tran

Not sure if I know enough about the algorithm to help unfortunately. What happens if you try 1000 samples per iteration? Maybe @naesseth can reply?

@dusenberrymw @danijar @markvdw We may end up moving Bayesian Layers code as well so that we can start building out examples and documentation.

You're correct. The namespace should be `edward` if we choose to move.

Thanks for the PR. To keep this Dockerfile minimally relevant to Edward, can you remove autograd? (Though I love autograd.)

+1. Can you make the same changes to Dockerfile-gpu?

Have you tried embedding layers, e.g., https://www.tensorflow.org/guide/embedding ? This compresses one-hot encodings into a low-dimensional real-valued space (e.g., vocabulary_size=1024 to embedding_size=64). This hasn't typically been done for regression analyses, but...

Gonna reopen as it would be a great tutorial/example to have in TFP!

@dusenberrymw There's a minimal code snippet in the paper. It should be plug-and-play with any LSTM example you currently have.

Thanks for raising this! I'm not super familiar with the implementation so cc-ing @jereliu.

@jereliu Hi @batzner! Were you able to resolve this?