Orange Forum • View topic - classification tree

classification tree

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

classification tree

Postby AFA » Mon Jun 11, 2012 14:01

Hi all,
I got a question about the classification tree for prediction.
In classification tree, we can see a selection of "Attribute selection criterion", in the case of "Relief F", what is the meaning of "limit the number of reference examples?" What will be th effects of changing this number?
thank you for your help.

Re: classification tree

Postby Ales » Tue Jun 12, 2012 18:02

The parameter controls on how many data instances is the Relief F computed (the outer loop of the algorithm). If unchecked, it runs on all instances.
Small values for this parameter produce unstable estimates of the Relief F score.

Re: classification tree

Postby AFA » Wed Jun 13, 2012 9:37

Ales wrote:The parameter controls on how many data instances is the Relief F computed (the outer loop of the algorithm). If unchecked, it runs on all instances.
Small values for this parameter produce unstable estimates of the Relief F score.


thank you.
but in my test, test results of small values for this parameter were more stable than big values or all instances, for example, I had 150 instances in my test, the result of parameter 50 was more stable than the parameter of 100.
this question is quite a puzzle to me.


Return to Questions & Support



cron