I think there's a typo in the cross-validation FAQ:
|
- ELPD: The theoretical expected log pointwise predictive density for a new observations (or other exchangeable entity) (Eq 1 in @Vehtari+etal:PSIS-LOO:2017). One scenario when we could also actually observe this is if we would get infinite number of future observations from the same data generating mechanism. However, this expected value is valid also when thinking just about one future observation (other exchangeable entity). This can be computed given different data partitions. For simplicity the ELPD acronym is used also for expected log pointwise predictive probabilities for discrete models. |
where
for a new observations (or other exchangeable entity)
should probably read "for a new dataset", although I'm not sure if the aim was to emphasize the "exchangeable" part (in that case, "for a new dataset" might have to be adapted).