I used Genie to learn a network but the resulting network is to complicated. I want to remove the weakest arcs.
1) How can I get a listing of the arcs and their strengths?
2) How can I remove the arcs below a given threshold?
Many thanks
Marcos
PS. I am using Market Research data about the brand image that has high multicolinearity.
Simplifying a learned network
-
- Site Admin
- Posts: 430
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Simplifying a learned network
Marcos,Marcos wrote:I want to remove the weakest arcs.
1) How can I get a listing of the arcs and their strengths?
2) How can I remove the arcs below a given threshold?
I'm afraid there is no automatic way of removing arcs in an existing network in GeNIe. To do this, you will have to use SMILE and write a program that will accomplish what you intend to do.
What you can do in GeNIe is display "arc strengths" in a network (Network menu, Strength of influence or a corresponding toolbar button) and then remove arcs individually. In order to prevent these weak arcs from appearing in the future, you can forbid them through the "Background Knowledge ..." dialog in learning. You can also try manipulating the p value in the PC algorithm. Lower p values should also lead to removing weak influences. I hope this helps.
Cheers,
Marek
-
- Site Admin
- Posts: 430
- Joined: Tue Dec 11, 2007 4:24 pm
The best is probably the book by Peter Spirtes, Richard Scheines and Clark Glymour "Causation, Prediction, and Search".Marcos wrote:1. Where can I find more information on PC, Greedy TT and a comparison of the methods?
I'm not sure I understand what you want to do but I'm pretty sure you will have to do it yourself in SMILE.Marcos wrote: 2. Can I create a listing of the arc influence in the output?
Sure. Whenever you save a model, you save it in GeNIe/SMILE format. SMILE can read it.Marcos wrote: 3. Can I export a network from Genie to use it in SMILE?
I hope this helps.
Cheers,
Marek