3 ms·
It seems that the post is comparing a predictive distribution conditioned on N data points to one conditioned on N-1 data points. The latter is a biased estimat
by to-mi 3y ago
It seems that the post is comparing a predictive distribution conditioned on N data points to one conditioned on N-1 data points. The latter is a biased estimate of the former (e.g., https://users.aalto.fi/~ave/publications/VehtariLampinen_NC2002_preprint.pdf https://users.aalto.fi/~ave/publications/VehtariLampinen_NC2...)
- sieste 3y agoThis. Leave-one-out doesn't turn an in-sample prediction problem into an equivalent out-of-sample problem, but a slightly less skilful one.