. 


--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/a2f29103-222e-485b-86c4-d1196abafdccn%40googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/CAKbe0hr0L%3D1JbKohQ%2Bhc7abTf3cbsFPwf%2Byz%3DnWEdtHhAL9abg%40mail.gmail.com.
Thank you Perry and Daniel - this is extremely helpful. In the
meanwhile I did also compare the MCMC efficiency of RW_block
(using adaptFactorExponent=0.8) against AF_slice. It's based only
on 5 runs, but the results show that the latter was nearly an
order of magnitude more efficient (btw thanks for dedicating a
slot on this topic in the workshop!). As for your question
regarding bi-modality - both stan and the run based on
adaptFactorExponent set at 0.8 returned unimodal posteriors for
all random effects, so presumably the bi-modality was just the
chains stuck into different regions?
Enrico