Hi,
I was working with the greedy thick thinning method to get a network from the data and came across the following problem. In the learned network, there is an arc missing from B to C (that is, B is not a parent of C), however the conditional probability table for the node C, gives B and F to be its parents. Why? The same happens when one obtains the netowrk with the Essential graph search method.
I have attached the GTT learned network for your reference.
Thanks for your reply.
JTJ
Greedy thick thinning
Greedy thick thinning
- Attachments
-
- GTTLearnedNet.xdsl
- (6.35 KiB) Downloaded 362 times
-
- Site Admin
- Posts: 1417
- Joined: Mon Nov 26, 2007 5:51 pm
Re: Greedy thick thinning
It's just an graphical artifact. GeNIe runs very simple graph layout algorithm (parent ordering) after GTT, which places nodes in lines. What you perceive as A->C arc is really B->C with A in between. Nodes are always drawn above arcs in GeNIe, so A simply obscures part of the A->C link.
If you move C manually you'll see that B and F are its parents. You can also use spring embedder algorithm from Layout|Graph Layout menu to re-layout the networj.
If you move C manually you'll see that B and F are its parents. You can also use spring embedder algorithm from Layout|Graph Layout menu to re-layout the networj.