search

lanl / SEPIA

Simulation-Enabled Prediction, Inference, and Analysis: physics-informed statistical learning.
Other
29 stars 6 forks source link

Setting up uniform priors for theta #50

Open teja781 opened 1 year ago

teja781 commented 1 year ago

Hello,

After reading the docs, I am still unsure about how to setup uniform priors for theta parameters and was hoping someone could help me out.

From what I understand, priors are set using "SepiaPrior" class, but since theta parameters are transformed, I think the bounds specified as part of the prior are for the transformed theta(? please correct me if I am wrong about this).

My question is how do I set up my theta priors with native space bounds?

Thank you Krishna

jgattiker commented 1 year ago

Hi, sorry for the long delay. The reason for the transformation is so the priors are defined on a standard domain, and then the priors do not need to be transformed. The domain of theta can then be taken to be [0,1] - which will either represent the range of the simulation dataset (default), or the range specified by the user during setup. The theta prior is a (truncated) normal. Uniform can be implemented, effectively, by making the variance of the prior very large. Other than that, we do have an issue to allow custom priors through function references, and maybe we'll get to that someday ...