Compilation error when sample_prior=TRUE

[eipi10]

I've been working with ZOIB regression models in brms and decided to try the ordered beta model after seeing your posts on the Stan forum and browsing through your paper (which I hope to read in more detail). I included sample_prior=TRUE as an argument in the ordbetareg function, which caused a compilation error. I'm not sure if I should have expected sample_prior_TRUE to work, but I wanted to bring this to your attention in case it's a bug. When I removed the argument, the sampling completed as expected. Now I just have to spend some time getting a better understanding of the model. Thanks for making this model and package available!

Below is a reproducible example using the pew data from the Vignette.

ord_fit_mean <- ordbetareg(formula=therm ~ mo(education)*mo(income) + (1|region), 
                           data=model_data,
                           cores=2,chains=2,iter=1000,
                           sample_prior=TRUE,
                           refresh=0)

Compiling Stan program...
SYNTAX ERROR, MESSAGE(S) FROM PARSER:
No matches for: 

  induced_dirichlet_rng(vector, int)

Function induced_dirichlet_rng not found.
 error in 'model165e36ef63963_4801811fdbe1356d482a6c93380fb357' at line 148, column 55
  -------------------------------------------------
   146:   real prior_phi = exponential_rng(0.1);
   147:   real prior_sd_1 = student_t_rng(3,0,2.5);
   148:   real prior_thresh = induced_dirichlet_rng([1,1,1]',0);
                                                              ^
   149:   // use rejection sampling for truncated priors
  -------------------------------------------------

Error in stanc(file = file, model_code = model_code, model_name = model_name,  : 
  failed to parse Stan model '4801811fdbe1356d482a6c93380fb357' due to the above error.
In addition: Warning message:
Rows containing NAs were excluded from the model. 

[saudiwin]

sorry about that! the model is using a custom prior and I didn't include an RNG function to allow it to sample from the prior. I'll try to update the package soon-ish. Would be a great feature to have for sure.

[saudiwin]

Unfortunately, this seems to be an underlying issue in brms, so it might take a while to resolve.

[saudiwin]

I made some changes to how the priors on the cutpoints are calculated, so it should work now. I also found a bug related to that calculation which could cause divergent transitions (!!), so on the whole this was a useful exercise. Try installing directly from github (remotes::install_github('saudiwin/ordbetareg_pack')) to use the sample_prior option.

[eipi10]

Great. I will install the updated package. Thanks!