Add robustness to PMCMC steps

nimble-dev / nimbleSMC

Sequential Monte Carlo methods for NIMBLE

2 stars 2 forks source link

Add robustness to PMCMC steps #13

Open perrydv opened 1 month ago

perrydv commented 1 month ago

PMCMC could be updated to:

check on valid priors before calling the particle filter likelihood approximation, and
check for invalid results from the particle filter and reject the proposed parameters.

paciorek commented 3 weeks ago

The first item is in @danielturek PR from December.

@perrydv regarding the second time, were you thinking of checking for NA/NaN/non-finite likelihood estimates?

perrydv commented 3 weeks ago

@paciorek I took a look at the second point and I think no change is necessary. The decide function that ultimately returns the logical for the accept/reject outcome checks for is.na (which catches both NA and NaN), so that should cover the issue already.