Orange Forum • View topic - Equal frequency discretization needs class labeled data sets

Equal frequency discretization needs class labeled data sets

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

Equal frequency discretization needs class labeled data sets

Postby sreastman » Tue Feb 05, 2013 15:49

I was refering to the documentation on Discretization (http://orange.biolab.si/docs/latest/ref ... etization/) when I noticed the note toward the bottom of the page:

Entropy-based and bi-modal discretization require class-labeled data sets.

I took this to mean the existence of a class column is not needed for other methods of discretization such as equal frequency. After a lot of debugging, I discovered it was, at least for equal frequency.

It would be helpful to add equal frequency into the note above.

Steve

Re: Equal frequency discretization needs class labeled data

Postby Ales » Tue Feb 05, 2013 17:17

sreastman wrote:I took this to mean the existence of a class column is not needed for other methods of discretization such as equal frequency. After a lot of debugging, I discovered it was, at least for equal frequency.
Equal frequency does not require class labeled data.
However the DiscretizeTable class has a bug preventing it from working with datasets with no class. At least if the clean parameter is True, if you set it to False then it works as expected for equal frequency discretization. We will fix this.

Re: Equal frequency discretization needs class labeled data

Postby sreastman » Wed Feb 06, 2013 13:26

Appreciate the feedback.

Thanks--Steve


Return to Questions & Support



cron