In fitting, when evaluating the posterior I clip aggressively using only "static" priors (exact line here). This appears to cause problems when, for instance, the distance prior changes dramatically. I should look into whether evaluating the prior for the MLE values for (s, A, R) over a chunk of the grid actually fixes this and doesn't dramatically increase the runtime.
Thanks to @catherinezucker for tracking this failure mode down!
In
fitting
, when evaluating the posterior I clip aggressively using only "static" priors (exact line here). This appears to cause problems when, for instance, the distance prior changes dramatically. I should look into whether evaluating the prior for the MLE values for(s, A, R)
over a chunk of the grid actually fixes this and doesn't dramatically increase the runtime.Thanks to @catherinezucker for tracking this failure mode down!