I was wondering whether you could give a summary of the current and recommended methods to derive the model likelihood and perform model comparison?
What I understood:
- LOO is recommended, but is apparently not available for SMC. I haven’t tried LOO yet but is it already available for PyMC3?
- The SMC marginal likelihood is still documented, can we use it safely? For the marginal likelihood, it’s not clear to me how to deal with the prior probability, does PyMC3 have a method to calculate it?