Rstan meanfield produces a parameter called lp__.1


#1

When running variational inference (meanfield) via rstan, it seems to produce a strange parameter called lp__.1 in addition to the ordinary lp__ parameter. A user reported this issue on the brms github page (https://github.com/paul-buerkner/brms/issues/387) and I want to make sure whether this is intentional before I try to fix it from the brms side.


#2

It is probably a bug in rstan.


#3

Although that “ordinary” lp__ is currently just zeros when using advi. In the future there will be lp__ and lq__ with proper values.


#4

My current problem with lp__.1 (whereever this comes from) is that it appears in stanfit@sim$samples, but not in stanfit@sim$fnames_oi.