Thanks for your clear explanation. I think better to leave it as it is. I hope this post would be useful to others who also have this confusion. Perhaps, if an explanation has been documented in the manual in detail, would be great to leave a pointer here.
Yes, I’ve been struggling with this for some time now. Though not showing in the cmdstan report is that I ran warmup=800, which is about 5 times of the sampling (150). My problem is that if I don’t run so many or if I set delta below 0.99, I easily get divergences.
I use 90,000 simulated data points. If I don’t use so much, it is hard to get estimates reasonably close to the true parameter values given the model complexity. The run time here is kind of the lowest achievable by me as a novice model builder. I have already used metric=dense_e, thanks to @bbbales2’s suggestion. Otherwise, the run time is over twice longer and also the N_Eff is less than one-tenth here given the short chain of the sampling stage.
May I ask: Given the data size, is the run time still unreasonable long compared to complex models you have seen? Is there anything a novice user can still try to see if there’s a quick fix in the slow run time?