Orange Forum • View topic - selecting features in Orange Canvas

selecting features in Orange Canvas

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

selecting features in Orange Canvas

Postby john » Tue Jun 14, 2005 17:56

The visual programming environment (Orange Canvas) looks great. Are there any plans to incorporate the feature subset selection module (orngFSS) into it?

Thanks,

John

Postby Blaz » Tue Jun 14, 2005 20:31

John, there's already a prototype for it on CVS called Feature Subset Selection (although only for selecting top N features). We are now working in polishing up other widgets, should get to this one within some reasonable time (but not in June).

Postby Guest » Tue Jun 14, 2005 21:25

Blaz,

I just downloaded the daily snapshot for windows. When I start up Orange Canvas, I don't see anything called Feature Subset Selection. Am I going about this correctly?

Thanks,

John

Postby Blaz » Wed Jun 15, 2005 14:49

You probably forgot to run "Rebuild Widgets Registry" (from Options menu in Orange Canvas). Withouth rebuilding the registry, adding the files in the widgets directories will not change the list of widgets in Canvas.

After you update Ornge with CVS, the FSS widget is in orange/OrangeWidgets/Data/OWFeatureSubsetSelection.py.

Postby john » Wed Jun 15, 2005 18:03

Blaz,

Thanks, that worked.

I'm now able to create this schema:

File ---- Feature Subset Selection ---- Test Learners
|
Naive Bayes ------

which will select the 10 best features and then test Naive Bayes via cross validation, using just those 10 features.

However, the feature subset selection needs to be inside the cross validation loop. In my schema above, I chose the 10 best features using *all* the data. Instead, the 10 best features should be chosen for each cross validation fold.

I can get this functionality when scripting, by using orngFSS.FilteredLearner. Is there a way to do it visually when using Orange Canvas?

Thanks,

John

Postby Blaz » Wed Jun 15, 2005 20:41

Not yet. The concept we have is that you will be able to add any number of preprocessors to the learners (which will essentially be wrappers around the learners). FSS, discretization, and constructive induction are examples for these. But it will take some time to get this to widgets, although, after your writing, we can prioritize this task.

[as you also write, all of these is simply doable in scripting, to put it in widgets requires some more work. our priority for the next two/three weeks is to polish-up the existing widgets and remove as much bugs as possible, we can focus on the ideas as described above only after orange 1.0 is released]


Return to Questions & Support



cron