contributed model repository

[Bob Carpenter]

On 3/4/12 10:43 PM, Andrew Gelman wrote:
> Hi Bob. We should be collecting a list of examples that can't be fit
> in Bugs, Jags, etc. Here's an example that might be worth looking
> into, ...

I'd like to get some mechanism for users to contribute
models, data, and short writeups (like the BUGS examples).

Do we want to vet them? I don't want to start the implicit
journal of Stan models and wind up with tons of editorial
and model-implementation work. But then I don't want any old
thing to get thrown into the example bin. Any ideas?

I took a quick look at the paper behind the example Andrew
was talking about:

Lockwood, J.R. et al. 2007. Bayesian methods
for scalable multivariate value-added assessment.
J. Ed. and Behav. Stat.

The authors say:

---------------
We implemented this model in WinBUGS (Spiegelhalter, Thomas, & Best,
1999), free software for Bayesian model fitting available at
http://www.mrc-bsu .cam.ac.uk/bugs/welcome.shtml. The WinBUGS
implementation was sufficiently fast to estimate the variable
persistence model with small data sets and the com- plete persistence
model with small or large data sets. However, fitting the vari- able
persistence model with large data sets or multiple subjects was
prohibitively slow. Therefore, we developed an MCMC algorithm (specified
in detail in the appendix) for sampling the posterior distribution and
programmed it in C, re- sulting in an efficient and scalable
implementation of the multisubject variable persistence model. We use
this algorithm for all results reported in this article.
---------------

The appendix has their own sampler, which works on the residuals
in the GLM setting (is this what the GLM module of JAGS does -- I
thought that was more about aggregating the sufficieint statistics).

This example is a nice example of what we'd like Stan to be able
to handle out of the box. But we don't do anything clever for
sparsity (yet, anyway).

- Bob

[Andrew Gelman]

I don't know about the big repository, I agree that (a) it would be good to have a Stan list like the Bugs list where people can help each other, and (b) I don't want to be vetting this. One approach would be to have some sort of standard format (analogous to the requirments for an R package) so that models could be checked automatically. Not a priority, though.

What I was more immediately thinking about was a set of examples for the papers and grant submissions, examples of real problems that Stan can solve but other approaches could not easily solve.

The hope with Lockwood's example is that it would run ok in Stan, either straight out of the box or if necessary with some customized code (which would then be there for future stan models)

A