Simple Stan test case

[dmenne]

library(ggmcmc)
stanmodelcode <- "
data {
int<lower=0> N;
real y[N];
} 
parameters {
real mu;
} 
model {
mu ~ normal(0, 10);
y ~ normal(mu, 1); 
} 
"

y <- rnorm(20) 
dat <- list(N = 20, y = y); 
fit <- stan(model_code = stanmodelcode, data = dat, chains = 3)
#### Another bug in ggmcmc: this dependency is missing. 
# Things should work without the next line
library(coda) # Bit this one gives an warning which should be avoided by using namespaces
s = ggs(fit)
#Error in 1:n.samples : argument of length 0

[dmenne]

Add-on to the above: once this works (my error?), try chains = 1 to reproduce #13

[jucor]

Regarding the coda dependency and the warning, this is fixed in Pull-Request #19.

[jucor]

Found the problem with gs(), fixed it in #20 : it happens when your model only has one variable. Until @xfim can merge in the official repo, feel free to use my master branch.

[xfim]

As for the first comment by @dmenne "#### Another bug in ggmcmc: this dependency is missing.". If you refer to the "rstan" package, it is not possible to add it as a dependency, because rstan is not in CRAN (see https://groups.google.com/forum/#!msg/stan-users/LMMxkuMinag/3fAgH3yjm0wJ).

[dmenne]

Correct, I had forgotten that. Maybe put that into the docs.

[jucor]

Good point! Do you know why isn't rstan in CRAN and why "you can't count on RStan being on CRAN any time soon" (to quote STAN's developer in your link above)?

[dmenne]

One reason is that CRAN is too limited when it comes to running larger test suites. CRAN is a horrible dinosaur; time for mammalian revolution.

[xfim]

No, @jucor, I don't know the reason.

[xfim]

rstan not being into CRAN added into the documentation of ggs by commit #e5194ec28. This should solve the issue.