Orange Forum • View topic - Classification Tree Data Setup

Classification Tree Data Setup

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

Classification Tree Data Setup

Postby spock » Mon Apr 15, 2013 18:08

Hi,

After reviewing the doc & trying on my own before that, I'm needing a question answered to what may be an intuitive answer for getting the "Classification Tree" widget to recognize the dataset & variable selection.

* Note, I've already successfully profiled the dataset (79k rows, 19 attributes) using the "Attribute Statistics" & "Distributions" widgets

To create a classification tree, using a dataset which has (4) levels of membership (gold, silver, bronze, & non-member), I'm trying to determine where within the "Select Data" or "Data Table" widgets I can select the field that is the dependent variable.

* I've already tried the "Classification Tree" as a child node from the "Select Data" or "Data Table" widgets, but am not getting any tree output nor any indication the process is started as the red stop sign button now occurs above the "Classification Tree" widget on the Orange workflow scheme.

Please advise as I really look forward to using Orange alot for visually based analysis & subsequently generating Python based workflow scripts.

Thanks.

Re: Classification Tree Data Setup

Postby spock » Mon Apr 15, 2013 18:15

Btw, of those (19) attributes, 14 are discrete & 5 are continuous, none are seen by Orange as string.

Re: Classification Tree Data Setup

Postby Ales » Mon Apr 15, 2013 18:49

spock wrote: I'm trying to determine where within the "Select Data" or "Data Table" widgets I can select the field that is the dependent variable.
Use the 'Select Attributes' widget to select the dependent variable (drag the variable to the 'Class' box).

Re: Classification Tree Data Setup

Postby spock » Mon Apr 15, 2013 22:54

Thanks for the assist, feeling pretty foolish I didn't try that one before ;->

Re: Classification Tree Data Setup

Postby spock » Tue Apr 16, 2013 0:27

Having used CHAID extensively in the past, I'm used to a methodology where I need to conduct exploratory data analysis initially for descriptive purposes on my dataset using a tree display or ruleset where I can manually *choose* which attributes of interest in common for each of the (4) unique values of my dependent variable.

* It appears from the results (or lack thereof) I may be violating the intended use case or procedural workflow for Orange.

Alternatively, based on how the Orange widgets assume input, would the "Interactive Tree Builder", "CN2" widgets, or "Associate Rules" be a better choice?

Thanks.


Return to Questions & Support