Closed qminor closed 2 years ago
Hi @qminor.
It means that your likelihood has the property that it does not return the same result if you call it again at a later time with the same inputs, which is a requirement if you want to perform slice sampling.
This can occur if there is a deliberate (unseeded) random element to your likelihood, or if there is a bug in ones code (such as accessing of deallocated memory, which would introduce unintentional randomness). In my experience it is not unusual for slice sampling to pick up bugs that have previously gone unnoticed (since this kind of nondeterminism doesn't affect Metropolis Hastings algorithms or MultiNest).
I ran a job that had to be stopped after 48 hours, but when I resumed it, I got the error "non-deterministic likelihood". What does this mean, and why would it happen upon resuming?
Thanks, Quinn