example model package coming?

[Sebastian Weber]

Hi!

From what I have seen an example models R package is planned - is that right?

I understood that this package pretty much includes all those very standard examples which are around, i.e. ARM and WinBUGS manual examples. Would it be possible to include the "original" BUGS files along with the Stan model codes? Reasons to do this:

- People know the BUGS stuff and can then easily compare what changes when going from BUGS to Stan.

- Also: WinBUGS is the golden reference when it comes to Bayesian statistics (at least this is its reputation in many fields). Hence, having the BUGS code nearby makes it easy for everyone to quickly cross-check that BUGS gives the exact same answer (whenever the model is sound, which those inverse-gamma priors may defeat).

Also: When is the example-models planned to see the light of this world?

Many thanks!

Best,

Sebastian

[Ben Goodrich]

On Sunday, March 29, 2015 at 7:44:08 AM UTC-4, Sebastian Weber wrote:

From what I have seen an example models R package is planned - is that right?

Was planned. But the plan was rejected by CRAN. The rstan 2.6.3 package includes the old stan_demo() function which now asks the user whether to download the example-models repository from GItHub.
 

I understood that this package pretty much includes all those very standard examples which are around, i.e. ARM and WinBUGS manual examples. Would it be possible to include the "original" BUGS files along with the Stan model codes? Reasons to do this:

The original is included in a URL at the top of a BUGS example, or you can install an R package that has them from

https://r-forge.r-project.org/R/?group_id=882

Ben

[Bob Carpenter]

We really need to revise the BUGS models. What we absolutely
don't want to do is stick to our literal translations of the
BUGS models, priors and all, because it doesn't match our recommended
practices for either coding in Stan or Bayesian modeling.

I understand we have a lot of that now, but it's leading
people to write bad Stan models.

I'll let Michael weigh on on cross-checking we get the same answers.
I believe they all match to the level you care about if you run each
for enough iterations.

- Bob

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

[Sebastian Weber]

Hi Bob!

I fully agree that best practices has changed, but still, I think that a direct translation from the BUGS models to Stan is valuable. As a compromise you could point out those translations for the ARM models only - there things are closer to what you recommend by now (no inverse-gamma problems).

In any case, I have seen many people being very shy to move to Stan while using WinBUGS is totally ok. This crowd is best taken over to Stan in little steps.

@ Ben: Too bad the example models package has been declined. Just one question: Where is the StanHeaders repository tracked or is it directly derived from the main Stan C++ repository?

Best,

Sebastian

PS: I am really looking forward to that 2.6.3 rstan with the function magic stuff! This is more than cool.

[Ben Goodrich]

On Monday, March 30, 2015 at 5:15:45 AM UTC-4, Sebastian Weber wrote:

@ Ben: Too bad the example models package has been declined. Just one question: Where is the StanHeaders repository tracked or is it directly derived from the main Stan C++ repository?

It is a subrepo of the rstan repo that tracks the Stan library

https://github.com/stan-dev/rstan/tree/develop/StanHeaders

Ben

[Aki Vehtari]

I think that "a direct translation from the BUGS models to Stan" may be useful, but the models should have BIG WARNING LABEL that the models and priors may be really bad.

If people google models, they should find models and priors based on the latest best practices. There should be a page explaining the generic changes made to the models.

I'm willing to help to update BUGS models, but maybe there should first be the page providing the guidelines for updating the BUGS models, and then any volunteer could use those guidelines to update some model and submit it for further checking by the Stan authorities.

Aki