Orange Forum • View topic - Rank Widget

Rank Widget

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

Rank Widget

Postby Andrew » Wed Mar 13, 2013 22:06

Hi,

I'm trying to use the Rank Widget as described here http://orange.biolab.si/docs/latest/widgets/rst/data/rank/#rank. Specifically, I'm routing 70% of the data to training, which feeds into the Ranker. In the ranker I select the top 40 features. A Random Forest widget receives the Reduced Data output and outputs a Learner to a Test Learners widget. Unfortunately, when I test the widget on the "Remaining Data" (the other 30% of the data output by the Data Sampler), I get a Python error which crashes the application.

I can see how this could be a problem: The Random Forest is trained on 40 attributes, but is tested on 87 attributes, however, this seems to be okay in the above-cited documentation.

If the test data has to have exactly the same attributes as the train data in "Test Learners", I'm not seeing a good way to do this on the canvas. Having used the "Rank" widget to select a certain subset of attributes from my training data, I don't see how to easily select those same attributes for my test data.

Thanks,
Andrew

Re: Rank Widget

Postby Ales » Fri Mar 15, 2013 13:05

This is (was) a bug in the Random Forest Learner. I have committed a fix so the problem should be gone in tomorrows snapshot.

Andrew wrote:If the test data has to have exactly the same attributes as the train data in "Test Learners", I'm not seeing a good way to do this on the canvas. Having used the "Rank" widget to select a certain subset of attributes from my training data, I don't see how to easily select those same attributes for my test data.

With the fix in place this should not be necessary, but there is a 'Translate Domain' widget in the 'Prototypes' Category (see this for instructions to enable it)


Return to Questions & Support