search

ReactiveBayes / RxInfer.jl

Julia package for automated Bayesian inference on a factor graph with reactive message passing
MIT License
260 stars 24 forks source link

Predictive posterior distributions #58

Closed caxelrud closed 12 months ago

caxelrud commented 2 years ago

Is it possible to generate predictive posterior distribution after posteriors are available?

bvdmitri commented 2 years ago

Hey @caxelrud!

Unfortunately, at the moment ReactiveMP.jl doesn't provide predictive posterior out of the box. We plan to add this feature in the future. However, it should be possible to obtain the predictive posterior manually. For the different problems, you'd have a different solution to how to do that. If you provide a more specific example, we'll be glad to help you out.

But generally speaking, you can try to initialise the same model you've used for inference, but replace uninformative priors with the inferred posterior distributions to generate predictive posterior distribution. You will probably also need to replace observations with uninformative distributions, e.g. Gaussians with huge variance.

svilupp commented 2 years ago

+1

albertpod commented 1 year ago

@bvdmitri, I will put this issue on hold until @ismailsenoz and I decide on how to incorporate this functionality generically. Some functionality is implemented in the dev-predict branch.