Hacker News new | past | comments | ask | show | jobs | submit login

Makes sense - it seems like this basically takes an idea that’s already applied frequently (fit a set of replicate models to prior predictive simulations) and adds a simple to apply comparison metric (data averaged posterior distribution to prior distribution). Usually I’ve just tried to quantify the performance of models fit to prior sims using ad-hoc measures, like posterior interval coverage, bias, and mse. This approach seems like a more Bayesian way to evaluate Bayesian inference.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact