rxmc.model_comparison.log_posterior_predictive

rxmc.model_comparison.log_posterior_predictive(logp_samples, logw=None) → float

log E_post[p(y_held | theta)]: the joint log posterior predictive density of the held-out block.

A log-mean-exp over posterior samples of heldout_log_predictive() values; pass logw (unnormalised log importance weights, e.g. nested-sampling logwt) for weighted samples. This is the joint predictive of the whole held-out block, not the pointwise-summed elpd of Vehtari et al.; divide by the number of held-out points for a per-point score.