Loo_compare elpd_diff output producing NaN

General
, ,
1

Hi all,

I am comparing models with the loo_compare function where I’ve used a kfold criterium.
Can anyone suggest why I am getting the following NaN elpd_diff and se_diff output.

              elpd_diff se_diff

sprfit2t 0.0 0.0
sprcrop_k30t -10.9 4.9
crop_spr3t NaN NA
crop_spr4t NaN NA

The two models with the NaN issue are as follows:

bf3<-bf(Total ~         b0 +
          b1 * (dist_3) +
          b2 * lambda * log1p_exp((dist_3 - omega)/lambda) +
          b3 * inv_logit((dist_3 - omega)/lambda),
        b0 + b1 + b2 + b3 + lambda + omega ~ 1,
        nl = TRUE)
crop_spr3<-brm(bf3, data = sprcrop_1.df, prior = bprior3,iter=26000,family="hurdle_gamma", control = list(adapt_delta=0.9999, max_treedepth = 14), core=6)

and 

bf4<-bf(Total ~         b0 +
          b1 * (dist_3) +
          b2 * lambda * log1p_exp((dist_3)/lambda) +
          b3 * inv_logit((dist_3)/lambda),
        b0 + b1 + b2 + b3 + lambda ~ 1,
        nl = TRUE)
crop_spr4<-brm(bf4, data = sprcrop_1.df, prior = bprior4,iter=4000,family="hurdle_gamma",control = list(adapt_delta=0.9999, max_treedepth = 12), core=4)

thank you in advance
2

Can you post the loo output for these two models in addition of the comparison result?

btw, you seem to have some big problem with your model as you have adapt_delta=0.9999. It rarely is sensible to have that larger than 0.99

3

Thanks for responding @avehari.

I’ve actually moved to simpler model design, so this is no longer an issue