How to use the posterior distribution of one model as a prior distribution for another model

9

Hi @Yamaguchi,

this example was actually a question (unanswered), I do not know whether it works or not. The point is, apparently you have the choice between:

  • have an empirical approximation of a distribution in the 1-D case (using Interpolated), but if with more than one dimension you loose the correlation (so this is generally not acceptable).
  • use a normal approximation of the n-D distribution, which is also not acceptable in general (e.g. one of your variables has a long tail…)

So this is pretty unsatisfying. @jessegrabowski pointed me to this histogram_approximation that looks promising, but we’d need someone with knowledge of the methods to comment on whether it would work or not (or just try and test it). Maybe @ricardoV94 ?

1 Like