pwinskill
commented
1 year ago This will now come under the $tuning() phase.
This will allow the user to set a starting number of rungs and a maximum number of rungs. The R side will then jump in and out of the MCMC, trying to estimate big Lambda and cobined that with a target average acceptance rate to calculate the required (min) number of rungs
Currently challenging and time consuming to calibrate the number and positions of temperature rungs. This should be automated. See: https://www.tandfonline.com/doi/abs/10.1080/10618600.2013.778779
Having too many rungs may result in disappearance of probability mass from some of the modes, see: https://doi.org/10.1214/08-AAP555 https://doi.org/10.1214/EJP.v14-638