Orange Forum • View topic - VisualProgramming: training classifiers for predictions

VisualProgramming: training classifiers for predictions

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

VisualProgramming: training classifiers for predictions

Postby Brad » Sat Feb 24, 2007 2:07

Hi

apologies if am a naive and frustrated acolyte. :oops:

I want to train e.g Knn from one a set of objects from one file and then predict for different object sets from another file(s).

I connect File to Knn to Predictions and File(2) to Predictions and get predictions if File = File(2).
If they're not the same then the behaviour appears flakey.

Have I got it all wrong? :(


Brad

Postby Blaz » Sat Feb 24, 2007 8:16

brad, you have to tell orange that although the files contain different attributes, those that bear the same name are indeed the same attributes. by default, orange would do this only if the collection of attributes in a file (called a domain) is entirely the same.

doing this is quite simple. you load one file with File widget, and the other one with Extended File widget which gets a domain descriptor from File. See http://www.ailab.si/orange/doc/widgets/ ... edFile.htm for description and example schema.

The schema that you need is then:

Image

For a test, I loaded a file golf1.tab in File, and golf2.tab in Extended File.
Last edited by Blaz on Sat Feb 24, 2007 14:53, edited 1 time in total.

Excellent help

Postby Brad » Sat Feb 24, 2007 14:30

:oops: :oops: RTFM basically :oops: :oops:
Thanks, that works like a dream

Brad


Return to Questions & Support



cron