Aggregate PSIS-LOO summaries can be read together with observation-level predictive contributions and influence diagnostics.
atlas <- create_loo_influence_atlas(
loo_result,
data = fit$specification$prepared$data
)
loo_influence_summary(atlas$table)
loo_flagged_data(atlas$table)
plot_loo_pointwise_elpd(atlas$table)
plot_loo_pareto_vs_elpd(atlas$table)
plot_loo_influence_rank(atlas$table)Flagged observations request inspection and are never removed automatically.