Confidence/credible intervals?

The front end.
Post Reply
RoC
Posts: 1
Joined: Mon Nov 22, 2010 10:49 pm

Confidence/credible intervals?

Post by RoC »

Is there any way currently in GeNIE of calculating CIs for the probabilities derived from a network? If not, this would be a very useful function to add!
It's easy enough to calculate CIs for the probabilities for the individual nodes, but the combination has me stumped! Any help, please? :?:
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Confidence/credible intervals?

Post by marek [BayesFusion] »

Doesn't this rely entirely on the model, which is often subjective anyway? What is the meaning of the confidence interval in this case? That the modeler is, say, 95% confident that the posterior is within the calculated bounds? I'm curious how you calculate the confidence interval for single probabilities and what you use them for.

When you are dealing with a data set from which you are learning the network, the question seem to me of more interest, as the answer hinges on the data. Confidence intervals can be calculated using, for example, bootstrap approach. You can write a simple program based on SMILE that will do that (including combinations of nodes, I believe).

I hope this helps.
Cheers,

Marek
Post Reply