lprior not available if normalize = FALSE

[n-kall]

Currently, specifying normalize = FALSE will define the lprior in the model block, so it is not returned in the fit (resulting in priorsense giving an error). At the least, this could be added to the documentation (such as is done for bridge sampling).

Alternatively, maybe the lprior could be created in the generated quantities using the normalized *_lpdf functions in this case.

[n-kall]

We could also provide an error message pointing to normalize = TRUE in the priorsense helper functions when the lprior is not available

[paul-buerkner]

Thanks! Since the _lupdf functions are not allowed in the transformed parameters block for some reason, lprior would have to be regenerated in generated quantites but that seems like excessive effort. The speedup from using the unnormalized functions is usually small and I think that it is easiest to just have an informative error that points to lprior not being saved because normalize was FALSE. I will update this in the next patch of brms.

[paul-buerkner]

Fixed :-)