Conformal scores with respect to predictive density
pat-alt opened this issue Β· comments
Hi @CoryMcCartan! Cool package ππ½ I want to add support for Conformalized Bayes in my Julia package as outlined here (Section 2.4, pp 10-11). I just had a look at your code for some inspiration and want to make sure I understand correctly: it looks like you're conformalizing point predictions (drawn from the posterior), is that right? In other words, conformal scores are not computed with respect to the posterior predictive density as described in the linked tutorial.
Thanks in advance!
Hi! Currently I basically expand a normal posterior quantile interval by a scaling factor set with conformal inference. So I'm not yet doing the level-set-based thing described on p. 11 of the paper you sent. Looks like that's probably a superior method, though, so I'll probably implement it if I ever get around to it.
Thanks!