Model comparsion for model fit with optimisation

[jeffreypullin]

trafficstars

Currently we implement loo() and (as of today) waic() to calculate model comparison metrics on the models fit using rater. Both functions, however, only support models fit using MCMC. This is because the underlying implementations in the {loo} package only supports MCMC-fit models. Ideally, we would also provide some way to compare models fit using optimization.

In the models fit using optimization we do have access to the log-likelihood evaluated at the MAP parameter values - but not the log-likelihood evaluated at the MLE. Unfortunately, I can't find any information criterion style metrics which work for models optimizing the log-posterior so I'm not really sure how to approach this issue.

@dvukcevic