Orange Forum • View topic - CART Regression - disappointing performance

CART Regression - disappointing performance

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

CART Regression - disappointing performance

Postby WhoDeeKnee » Tue Dec 26, 2006 8:58

Hi,

I have been experimenting with orngTree and the domain10.py example. I modified it so that it works nicely and can choose the best, minimal set of attributes for discrete classification using TreeClassifiers. Attributes that I know to be bad are easily identified as bad and lead to trees with bad RMSE's, and I get good RMSE's with what I know to be my best imput attributes.

In my input file I specified that the "class" variable is continuous.

However, my problem is a regression problem.

To switch to regression, I simply replaced this line:
rt = orngTree.TreeLearner()
with this line:
rt = orngTree.TreeLearner(measure="retis", mForPruning=2, minExamples=20)

Now ALL my attributes perform the same - the trees induced have the same RMSE's. There is no difference in performance between practically random inputs and excellent inputs highly correlated with the output variable.

Did I do something wrong? Has anyone had any luck using the "retis" algorithm with Orange?

Thanks!

Return to Questions & Support



cron