Errors; Hurdle model in brms

Modeling
1

I would like to hurdle model for fish catch with brms packages.

ff <- read.csv(file.choose())
ff <- merge(f, n[,1:7], by = "code", all.x = TRUE) %>%
  filter(!is.na(sigma.BTM_rcp26_2005.5YS) & !is.na(sigma.S100_rcp26_2005.5Y.1) & !is.na(to.BTM_rcp26_2005.5YAVE.) & !is.na(to.S100_rcp26_2005.5YAVE))

ff.csv (66.5 KB)

fbmodel <- brm(total_2005 ~ Thickness.average + Depth.range + Depth.average + 
                                sigma.BTM_rcp26_2005.5YS + sigma.S100_rcp26_2005.5Y.1 +
                                to.BTM_rcp26_2005.5YAVE. + to.S100_rcp26_2005.5YAVE, 
                                offset(totalnet_2005) , 
                   data      = ff,
                   family    = hurdle_negbinomial(link = "log", link_shape = "log", link_hu = "logit"),
                   prior   = c(set_prior("normal(0,10)", class = "b")),
                   warmup    = 300,
                   iter      = 1000,
                   chains    = 4,
                   save_all_pars = TRUE,
                   backend = "cmdstanr",
                   control = list(adapt_delta = 0.99, max_treedepth = 15),
                   save_model = "fish_stanscript",
                   file = "fish")

but it returned the error;

 Chain 4 Rejecting initial value:
    Chain 4   Log probability evaluates to log(0), i.e. negative infinity.
    Chain 4   Stan can't start sampling from this initial value.
    Chain 4 Initialization between (-2, 2) failed after 100 attempts. 
    Chain 4  Try specifying initial values, reducing ranges of constrained values, or reparameterizing the model.

hurdle_negbionomial was not suitable for this data?

2

@marie_stan looks like a reasonable selection based on the data. Did you try making a simpler version of this model first (say with fewer predictors or some simulated data)? Also it looks like you don’t have any predictors of the hurdle component?

3

Thank you very much for the comment.

I changed into a simple setting (one variable considered) and added “hu ~” components.
It successfully seemed to work, but increasing any variables, again the same error came up.

fbmodel <- brm(bf(total_2005 ~ Depth.average,
                  hu ~ Depth.average),   # give 1 instead, it works
                   data      = ff,
                   family    = hurdle_negbinomial(link = "log", link_shape = "log", link_hu = "logit"),
                   prior   = c(set_prior("normal(0,10)", class = "b")),
                   warmup    = 500,
                   iter      = 1000,
                   chains    = 4,
                   save_all_pars = TRUE,
                   backend = "cmdstanr",
                   control = list(adapt_delta = 0.99, max_treedepth = 15),
                   save_model = "fish_stanscript",
                   file = "fish")

Thanks,

4

I fount it successfully worked.

fbmodel <- brm(bf(total_2005 ~ 
                   Thickness.average + Depth.range + 
                    Depth.average +
                    sigma.BTM_rcp26_2005.5YS + sigma.S100_rcp26_2005.5Y.1 +
                    to.BTM_rcp26_2005.5YAVE. + to.S100_rcp26_2005.5YAVE + 
                    offset(log(totalnet_2005)), 
                  hu ~
                    Thickness.average + Depth.range + 
                    Depth.average + 
                    sigma.BTM_rcp26_2005.5YS + sigma.S100_rcp26_2005.5Y.1 +
                    to.BTM_rcp26_2005.5YAVE. + to.S100_rcp26_2005.5YAVE ),  
                   data      = ff,
                   family    = hurdle_negbinomial(link = "log", link_shape = "log", link_hu = "logit"),
                   prior   = c(set_prior("normal(0,10)", class = "b")),
                   warmup    = 500,
                   iter      = 1000,
                   chains    = 4,
                   cores     = 4,
                   inits = 0, 
                   save_all_pars = TRUE,
                   backend = "cmdstanr",
                   control = list(adapt_delta = 0.99, max_treedepth = 15),
                   save_model = "fish_stanscript",
                   file = "fish")
5

Great, is it the offset(log(totalnet_2005)) that resolved the error? That variable has a lot of zeros in the data. Wouldn’t that cause those rows to be dropped?