Dustin Tran

A first step could be improving the currently minimal notebook with the various text and visualizations from Edward1's. A next step would be adding your own favorites. :)

i'll have a lot more time to contribute to this after friday's icml deadline as well. :) (for example, finally finishing up conjugacy)

There's [this on the Edward forum](https://discourse.edwardlib.org/t/a-simple-tensorflow-implementation-of-forward-backward/67). Conjugacy hasn't been merged in yet (#588), so that's what personally been delaying my thoughts on the matter. I haven't figured out how exact...

Two options for integration are by: (1) wrapping pyhsmm algs in a pyfunc, or more simply, (2) defining the parameters you'll learn with pyhsmm as `tf.placeholders` in the Edward model....

We have no support for CRFs other than to the extent that TensorFlow is a numerical library and we have features to perform inference given unnormalized target densities.

Not at the moment. I was working out RB as you pointed out, which turns out to be quite involved. Control variates are much easier to implement; I haven't looked...

I've never experimented with it so I'm not sure. pinging @franrruiz. I do think there's no one-size-fits-all solution—especially for discrete random variables, and whether it be score functions+control variates, g-rep,...

Thanks for working on this! Ping me whenever you'd like some feedback.

Great work! Will take a look soon. For questions I can answer immediately: > It looks to me like you don't explicitly sample-via-reparameterization from q(z) in, for instance, build_reparam_loss_and_gradients. Why...

Apologies for the delay! Busy for ICML stuff due this Friday. Maybe ping me this weekend? :)