Equation nodes and choice probabilities?

The front end.
Post Reply
a.s
Posts: 10
Joined: Thu May 10, 2012 10:44 pm

Equation nodes and choice probabilities?

Post by a.s »

Hi

I am modeling a class of students doing peer-assessment on each other.
Each student is modeled by a small BN, and all the sub-nets are connected depending on the assessments.
The teacher sets as evidence the student's choices and is guided through the correction.
This is done in Yap Prolog with CLPBN.

I would like to mimick the same process in Genie because of its graphic interface/editor.
I am able to build the peer-assessment network and all works fine, but I would like to guide the teacher by highlighting the next node (answer) to correct.
Instead than highlighting it, it would be OK to compute a function depending on a set of nodes and their current probabilities.

To this aim I am trying to use Equation nodes to compute either:
- the entropy of a node's probability distribution
- or the current probability of a given value for a node [i.e. the P(X=val1) ]

Is it possible to define an equation using the current probabilities for the values of a node? (e.g. to compute its entropy or the max prob.)
What is the correct syntax to read a node/value probability?
Is there some good example of networks using equation nodes?
Is it possible to use both Choice and Equation nodes in the same network?

Thanks a lot
A
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Equation nodes and choice probabilities?

Post by marek [BayesFusion] »

Hi A.,

If I understand you correctly, you would like to add a continuous (i.e., "equation") node to a discrete Bayesian network and then from it refer in the definition (i.e., equation) to the probability of one of its discrete parent nodes. Please correct me if I'm wrong.

If my understanding of what you want to do is correct, GeNIe does not allow this as of now. GeNIe will allow for simultaneous discrete and continuous nodes only when using the autodiscretization but will not allow arcs between discrete and continuous nodes and there is no function defined that refers to the probability of another node. We have put the developments of continuous nodes somewhat on hold but plan to come back to them at some point, so stay tuned. Sorry we can't help you at the moment.
Cheers,

Marek
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Equation nodes and choice probabilities?

Post by marek [BayesFusion] »

I have just realized that you could perhaps solve your problem using the diagnostic capabilities of GeNIe but I'm not 100% sure. If you designate your nodes of interest as diagnostic targets and the nodes that the teacher wants to focus on (the things to correct), GeNIe/SMILE will automatically calculate the cross-entropies between them in the diagnostic dialog. I hope you know what I am talking about? You can switch all this on by selecting "Enable Diagnosis" in the Network menu.
Cheers,

Marek
Post Reply