Orange Forum • View topic - Evaluation widgets for Regression Tree

Evaluation widgets for Regression Tree

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

Evaluation widgets for Regression Tree

Postby mccormackpj » Fri May 03, 2013 20:48

Firstly, I am a complete Orange novice - but that will become quite obvious very quickly!

I use CHAID and CART regularly in my direct marketing world and have successfully used AnswerTree and Statistica in the past. For example, a common task is to predict responsiveness to communications (categoric dependent variable) or first order value / net annual value (continuous dependent variable).

I'm trying to replicate these tasks in Orange Canvas. Although I have had some success with the Classification Tree, hooked up to Test Learners and a Confusion Matrix, ROC Analysis, for a categoric dependent variable, I am mightily stumped when trying to reproduce a CART-style regression tree!

Yes, the attributes / variables, both predictors and dependent, are all continuous. Yes, the tree appears to be generated and results are displayed as 'Evaluation Results' in the Test Learners, but no tree is visible in the Regression Tree Graph and all of the suggested Evaluation widgets (ROC, Lift) are displayed with the warning "xxx cannot be used for regression results".

Have I fundamentally misunderstood what Orange does with regression trees? Have I bust it? Or just my poor fuddled brain? :? I have tried to find examples and answers in the documentation, but can only find detailed references to scripting in Python rather than examples of how to build analyses with widgets in Orange Canvas. I had hoped that I could do all I wanted without resorting to scripting, but then maybe I am an idealist. :D

Could someone take pity and point me in the right direction? Or suggest some examples to work through?

In anticipation ..

Patrick

Re: Evaluation widgets for Regression Tree

Postby Ales » Wed May 08, 2013 12:12

mccormackpj wrote:Yes, the tree appears to be generated and results are displayed as 'Evaluation Results' in the Test Learners, but no tree is visible in the Regression Tree Graph

This should not be the case. Can you provide some more details about your workflow? Does any widget display an error or a warning icon in the canvas?

As an example: loading a 'housing.tab' file from the documentation datasets into the 'File' -> 'Regression Tree' -> 'Regression Tree Graph' scheme should displays the tree in the Regression Tree Graph. Can you test this on your installation?

mccormackpj wrote:and all of the suggested Evaluation widgets (ROC, Lift) are displayed with the warning "xxx cannot be used for regression results".
Yes, the ROC, Lift and Calibration widgets only work on classification tasks. Orange is lacking in this department.


Return to Questions & Support