PSIS-LOO of a Bayesian GLM

Leave-one-out expected log predictive density on the training data, by Pareto-smoothed importance sampling (elpd_loo()), with each observation's relative efficiency estimated from the chains. Its pointwise values feed stacking_weights().

Usage

bayes_loo(model)

Arguments

model

A bayes_glm_model.

Value

As elpd_loo().

Examples

d <- data.frame(claims = rep(c(1, 2, 3, 5), 10), x = rep(c(-1.5, -0.5, 0.5, 1.5), 10))
m <- bayes_glm_fit(claims ~ x, d, family = "poisson", chains = 2, tune = 300, draws = 300)
bayes_loo(m)$estimates
#>         elpd           se            p           ic 
#> -56.62809857   1.73930396   0.07649576 113.25619714