Quantcast

Attribute at the root of J48 tree different when pruning turned on/off

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Attribute at the root of J48 tree different when pruning turned on/off

maytalsaar
This post has NOT been accepted by the mailing list yet.
All:

For a given data set I found that WEKA's j48 chooses in a different attribute as the first split (at the root) when J48 is running with the pruning option turned on or off.

This should not be happening, unless there is some randomness in the process. Specifically, in the example
above j48 selected a numeric attribute when pruning was turned off. Hence I was wondering if there is any random element in the search of split point for numeric attributes, and the pseudo random number had something to do with the J48 parameters for which  the infoGain computed when pruning was turned off might be different than when it is turned on.

Otherwise, is there another explanation?

Thanks!
Maytal.
Loading...