I’ve got a model that is running slowly, and I’m wondering if I might be able to speed things up by scaling the parameters. I’ve seen several suggestions in this forum and the user’s manual to try to keep the unconstrained parameters on a unit scale, but I’m unsure how one should best go about to do this. I’ve extracted draws of the unconstrained parameters, and I’m attaching a histogam of their SD’s and means. The largest SD (.7) is about 180 times larger than the smallest (.0039).

If I want to scale the parameters optimally, could I simply identify those with the least variance and divide these by some constant before using them to increase their scale? Would this be best done using the transformed parameters block? Are there any examples or guides on how to approach this?

posteriorSDs.pdf (22.6 KB)