a few things variational

[Alp Kucukelbir]

hi all,

a few questions:

1. any update on the advi pull request? anything i can help with?

2. did we figure out what to do with documentation that doesn't fit into the manual? (e.g. see attached; these are helpful to understand the advi model-based tests.)

3. on the topic of documentation, here's what i had in mind: a very brief explanation of variational inference and advi in the manual + the interface details in the cmdstan manual. sound good?

4. is there any code for EP in Stan at this point? i'm genuinely interested in studying fullrank advi v. gaussian EP.

cheers
alp

[Daniel Lee]

Sorry, a bunch of us are putting some time into a grant with a deadline this week.

Aki has some EP code: https://github.com/gelman/ep-stan

I started working on a native implementation using the C++ api, but that didn't go very far.

Daniel

alp

--
You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Michael Betancourt]

Alp, throw the doc in https://github.com/stan-dev/blackboard

[Ido Rosen]

Should I be able to see that repo?  It's 404ing for me.  

[Alp Kucukelbir]

@dan: ok no worries. thanks for the pointer; i'll take a look.

@michael: it's also 404ing for me.

[Aki Vehtari]

https://github.com/gelman/ep-stan is public so you should see, but it's for "EP as a lifestyle paper" so it's not what Alp is looking for.

Aki

[Michael Betancourt]

Alp, try again.

[Alp Kucukelbir]

@aki: it's not exactly what i was looking for, but still interesting. i'm working with some similar hacks as i'm developing the stochastic version of advi. so this is helpful!

@michael: perfect. pushed! thank you.