Hi,
I am using sample_smc, with a blackbox likelihood. To calculate the pointwise log-likelihood, I am using a customDist with a logp function. This all works fine when using 1 core but fails for multiple cores, sigificantly slowing down the computation.
I have seen threads on GitHub regardig this issue:
- BUG: `pymc.sample_smc` fails with `pymc.CustomDist` · Issue #7224 · pymc-devs/pymc · GitHub
- Register the overloads added by CustomDist in worker processes by EliasRas · Pull Request #7241 · pymc-devs/pymc · GitHub
- Compile the functions needed by SMC before the worker processes are started by EliasRas · Pull Request #7472 · pymc-devs/pymc · GitHub
I tried reading through the posts but could not understand anything. Is there any work around for this issue? This occurs on both my Windows and MacOS machines.