I’m often in a situation where
loo.object <- loo(mymodel) returns a LOO object with 2 problematic observations, and calling
loo.object2 <- loo_moment_match(mymodel, loo = loo.object) gets rid of one of them but not both. The resulting loo.object2 still has 1 observation with an excessive pareto-k.
When this happens, is there any way to use
reloo = TRUE to “pick up where moment matching left off”, so that the model has to be refit only once (dealing with the single problematic observation still remaining) rather than twice, as happens when you call
loo(mymodel, reloo = TRUE)?