Orange Forum • View topic - Problem with Orange Canvas

Problem with Orange Canvas

A place to ask questions about methods in Orange and how they are used and other general support.

Postby Sam_Nav » Mon Mar 27, 2006 14:55

Martin,

Thank you very much for such a detailed reply!!

The reason I asked you this question was that even though I've 30 variables ( and that I know that at least ten of them are important), the rule length discovered by CN2 widget is at max 4. I would like it to make long rules, but albeit changing the settings, it's still keeps the rule length to 4. At the moment, I've got minimum coverage and max rule length both set to zero, so the rule discovery should be very exhaustive. It seems it isn't. Beam width could be a problem because I've got it set at 3, so I am going to try increasing it. I haven't tried weighted covering yet as well so I'll try it now as well. The default value for weighted covering is set to 0.9 so I'll play around with this number as well.

I believe that there are many dependant concepts in my dataset which makes it very interesting and ideal to start data mining. You've pointed out that in that case I should use weighted covering. Can you just explain what the figure (0.9) in weighted covering actually does and what do I need to set it to get very exhaustive search?

I really appreciate your help in this regard.

Thank you very much again.

Sam.

Postby Martin » Mon Apr 03, 2006 22:19

Sam,

I am sorry for a late reply but I have been travelling last week.

If the maximum length of induced rules is 4, then it means that combinations of four attributes are enough for description of a single concept. In case that you believe 10 attributes should be used, they can be used within differnet rules and I, then, would also try with weighted covering.

Weighted covering means that instead of removing covered examples by previous rules, examples' weight is multiplied with a selected number. At the beginning each example has weight 1. If covering factor is 0.9, then after covering examples once their weight will be 0.9, after twice 0.81, and so on.

Hope it helps,
Martin

Previous

Return to Questions & Support



cron