Expert elicitation for CPT's

The front end.
Post Reply
Lotte Yanore
Posts: 12
Joined: Thu Feb 25, 2021 3:23 pm

Expert elicitation for CPT's

Post by Lotte Yanore »

My name is Lotte Yanore. I recently started using GeNie as a part of my PhD research.

I am currently exploring methods used to reduce the burden on experts when eliciting CPT's. For example, we want to use the ranked node method and the Noisy-MAX gate.

My question is, to what extent does your software support the use of any of these methods.
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Expert elicitation for CPT's

Post by marek [BayesFusion] »

I'm afraid GeNIe does not (yet) have the ranked node method implemented.

GeNIe has Noisy-MAX gates and there is also algorithmic support for them, so models Noisy-MAX nodes will update faster (not that it is very needed -- SMILE is very very fast :-)). I assume you have seen the graphical elicitation tools in the definition of discrete nodes?

There are two indirect elicitation tools that I would like to point out:

(1) continuous nodes that are specified by an equation that can include practically any functions and any probability distribution; you can use them to derive CPTs -- you need to specify the discretization and then discretize the network or just copy the CPT from the discretization tab, and

(2) QGeNIe, which implements DeMorgan gates. QGeNIe models consist of all propositional nodes but models require n+m parameters, where n is the number of nodes and m is the number of arcs. QGeNIe models can be exported to GeNIe for further refinement.

I hope this helps,

Marek
Post Reply