Hi Ricardo, thank you for the quick response! You covered pretty much all of it.
point 1. I guess this is trivial and I only need to supply the log-likelihood, in the Gaussian case this is just the sum of the square of the differences with a minus sign in front, omitting the scaling.
point 2. OK, so when I call pm.sample and give the log-likelihood, the posteriors are computed by taking into account the priors, so proper bayesian way.
point 3. I tried running what you suggest, basically the following:
log_l = LogLike(log_likelihood, logging_df, sigma=1)
pm.Deterministic(“likelihood_term”, log_l(dips, azimuths))
pm.Potential(“likelihood”, log_l(dips, azimuths))
but I still do not get the likelihood term in my inference_data object…
point 4. I see, so I just need to omit the __interval suffix variables, got it!
Many thanks. Maybe if you can just elaborate a bit more on how to use both pm.Deterministic and pm.Potential together, if you have the time!