Child node not calculating probability

The front end.
Post Reply
afkkarpm
Posts: 2
Joined: Thu Apr 07, 2022 7:02 am

Child node not calculating probability

Post by afkkarpm »

Hello everyone!

We have run in to some problems with the software, which i cannot find the solution in the Genie Manual, that is why I would like to ask your help.

Our problem is, we built a complex network, with a lot of connections, but the child nodes that the others point to, doesn’t calculate the probability from the parent nodes. The hundreds of cases listed on the definitions tab are only filled with 50% probability, however the parent nodes are not 50%.

I attached an example, which is only one child node and its parent nodes from the complex network.

Thank you
Attachments
Example.png
Example.png (37.8 KiB) Viewed 1865 times
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Child node not calculating probability

Post by marek [BayesFusion] »

Your screen shot shows that the definition of the node Munkaterulet is trivial, i.e., none of the parents influences/makes any difference to the posterior marginal probability of the node. You can also see it in the color of the incoming arcs, which are all grey. This means that the node Munkaterulet is independent of all of its parents (and the arcs are not playing any role and are not necessary -- this is why they have a grey color). No matter what evidence you enter in the parents, the posterior here is going to be always 0.5/0.5.

You need to specify the conditional probabilities in the table that you showed in the screen shot (you can just enter numbers in there or use the graphical tools like pie chart or bar chart). These are probability distributions for each possible combination of the parent states. Once you have done this, you will see the arcs becoming darker and also the posterior probabilities will no longer be uniform. I suggest that you work through the "Hello GeNIe!" section in the manual -- you will see how to specify probabilities.
I hope this helps,

Marek
afkkarpm
Posts: 2
Joined: Thu Apr 07, 2022 7:02 am

Re: Child node not calculating probability

Post by afkkarpm »

Dear Marek,

thank you for the detailed answer. As I understand, i have to add the probabilities manually to the child node, but I still want to ask, is there a way the program can help calculate and fill the brackets? There are hundreds of cases in my network, it would help a lot.

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

Re: Child node not calculating probability

Post by marek [BayesFusion] »

This is a "million dollars question" :-). Populating CPTs (Conditional Probability Tables) is generally a laborious task. If you have data that you can learn from, it helps a lot -- the tables can be learned from data. If not, then manual labor, getting the numbers from expert knowledge. One thing that you should look carefully at are the so-called "canonical gates." Please look at the corresponding chapter in GeNIe manual. You will find a link to a tech report by Javier Diez and myself that reviews canonical gates. Noisy-OR and Noisy-MAX are most popular and are absolutely recommended. Another technique that I recommend is to use QGeNIe to produce the first version of your tables. QGeNIe uses a canonical model called DeMorgan. It requires one parameter per parent and one parameter per node. Once you create a qualitative model in QGeNIe, you can export it to GeNIe and have the first version of your table that you can refine further.
I hope this helps,

Marek
Post Reply