Now that I looked at the travis details, I think this might currently only be building for 0.4. I am interested in helping out in this repo, would you be looking to get this package ready for 0.5 and 0.6?

Thanks!

Unfortunately, I don't currently have much time to devote to this repo. Part of that is general busy-ness. Part is that I'm up in the air about what the library should be moving forward. Specifically:

1. I spent a lot of time writing macros to handle and assess conjugacy, but much of the VB literature has moved away from this and toward black box methods. Those are actually easier to code. I'm very tempted to remove a lot of the extra conjugacy code to slim down the library.
2. The work @ccxxxx has done on ADVI is very nice and would likely stay. The big update here would be to use ReverseDiff rather than ForwardDiff.
3. I'm also on the fence about whether something more like Edward, which is primarily a convenience layer on top of TensorFlow, might be more useful to the community.

If you're interested in contributing, which direction it takes should probably be informed by what use cases you have in mind. I'm happy to help review PRs and perhaps give you commit access, if you had a vision you wanted to implement.

On points 1 and 2 I think I definitely agree. I am trying to commit some to Edward at the moment because I think that framework really makes sense. And yeah, it looks like most of the action in the space right now is in things like BBVI.

If at some point you have a plan for the direction of the library, I would definitely be interested in helping, but I also don't think I have the time or knowledge to lead the direction.

Thanks for the reply!