Replicated data in new schools in stan

[Linas Mockus]

Hi,

Is it possible to implement predictive posterior simulations in stan instead of R?  I already have stan fit so it should be straightforward.

theta_rep <- array (NA, c(n_sims, J))

y_rep <- array (NA, c(n_sims, J))

for (s in 1:n_sims){

theta_rep[s,] <- rnorm (J, schools_sim$mu[s], schools_si

m$tau[s])

y_rep[s,] <- rnorm (J, theta_rep[s,], sigma)

}

Thank you,
Linas

[Bob Carpenter]

Yes and no. You can generate fake data in one of
two ways: using the generated quantities block or
parameters plus a model. The latter is more general,
but also more costly because it's MCMC, not just MC.

There are two problems, though. First, you have
to write the data generating model in Stan, which isn't
like the original model. You can often do this using the
parameters approach by swapping what's data and what's
parameters.

The second problem is that the output CSV format for
Stan can't easily be fed back into Stan. So if you then
want to iterate and fit again (as Andrew's always suggesting),
then it's not so easy.

We'll hopefully have a routine to do this soon, at which
point I'll write a chapter into the Stan manual on how to do
it from CmdStan.

- 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/groups/opt_out.