Re: Stan development

[Bob Carpenter]

[cc-ing stan-dev list; I just gave you (Ben) posting permission]

Ben:

What are you interested in doing? We can use help on both
modeling and C++ coding:

modeling: bring existing examples up to best practices for modeling and coding
add new examples in your area of expertise

C++ coding: always more functions to implement
always more doc to write --- we want API doc throughout
always more tests to be added
always outstanding issues to debug

Interface & ShinyStan coding: I'm not sure what's needed there

I'd say starting simple is the way to go then you can work up to something
more ambitious if you're interested in more substantial contributions. You
can look through our outstanding issues and see if there's anything you want
to tackle.

We have a meeting every Tuesday (I think they're going to move to
11 AM and maybe even move to Wednesday --- not sure). That's a mix
of modeling (or whatever else Andrew or Michael are thinking about) and
software.

- Bob

> On May 6, 2015, at 1:39 PM, Ben Lambert <ben.l...@some.ox.ac.uk> wrote:
>
> Hi,
>
> I would very much like to be a part of developing Stan; if I can contribute in any way, that'd be great.
>
> In terms of my background, I am a researcher at Oxford University in computational\mathematical biology. I have a reasonable amount of experience of developing in c++, and teach Bayesian statistics. I am also passionate about the use of open source software, as well as the use of the internet to spread ideas, and teach.
>
> I am willing to help in any way I can; whether it be basic development of software, or contributing to any supplementary documentation etc.
>
> Let me know if you need any further information, and I will do my best to provide it.
>
> Best,
>
> Ben Lambert

[Bob Carpenter]

> On May 7, 2015, at 10:27 AM, Ben Lambert <ben.l...@some.ox.ac.uk> wrote:
>
> Hi Bob,
>
> I don't know how you would feel with me working on the following:
>
> - Implementing the remaining BUGS examples into Stan.

I don't know that the ones we haven't coded are worth it.

What we really need is all the BUGS examples brought up to
our best practices in terms of modeling (mainly priors) and
coding (lots of issues there). We want them to be good examples
to give people.

There's also the issue of replicating all the BUGS examples
exactly as coded in BUGS. We'd ideally like to have both the
direct model translation and the "right" way to do it.

And in the best case, we'd like to get simulated data from
the model to ensure the software fits it.

And also the Gelman and Hill book's models which need a similar
going-over.

I'm forwarding you a recent discussion on what needs to be done
for a single model.

> - I am currently working on a meta-analysis of data with a focus on malaria. It is a really nice, rich data set, with lots of potential for hierarchical modelling. It would be great to eventually include this as an example.

That'd be great. We don't have a nice meta-analysis
example although I mention it briefly in the manual.

There's a branch for the manual if you want to add examples
or there's an example-models repo where we have our example
models (including the BUGS examples).

> - Beginning to do a bit of coding of Stan/RStan. Could start with perhaps looking into Boost's Chebyshev polynomials, and how these could be implemented.

See this closed issue:

https://github.com/stan-dev/stan/issues/37

Nothing ever made it as far as Stan.

- Bob