Combining several variable selection techniques in Enterprise Miner

[Tijl]

Hi there,

I'm using Enterprise Miner 5.3 and am trying to combine the results of several variable selection techniques into a larger pool of "good" variables. I'm using the Decision Tree, Regression and Variable Selection nodes to flag variables as Rejected. What I want is to combine all variables that are not rejected in either or all of the three nodes.

Originally I used the Drop node to drop all Rejected variables for each technique and then used the Merge node to combine the remaining variables into a pool to be used for further modeling. This seems to work fine, as evidenced by the succesful running of the Model Comparison node, however it all goes haywire when I try to combine some of my models with the Ensemble node. It then complains with the following warning (and ensuing errors): "WARNING: Apparent symbolic reference EM_SCORE_OUTPUT not resolved." (Log attached)

1) Does anybody have any idea what causes the error?

2) Am I doing the merging wrong? Is there a better way?

Thank you for your time,

Tijl Kindt

[JasonXin]

Hi, Tijl,

This TS note appears to address your issue with EM 5*x. There is a hot fix there. The root cause is the Merge Code.

Problem Note 19447: Incorrect Score Code generated from Merge node

Jason Xin

[Tijl]

Hmm... Thanks for the reply.

This webpage claims the bug was reported in 5.2 and should be fixed in 5.3, right? So in theory, I shouldn't be getting this error, right?

19447 - Incorrect Score Code generated from Merge node

[jwexler]

Hi at this point I would recommend working with SAS Tech Support.  They are an excellent resource for issues like this, and can provide 1:1 support and solutions. 

You can open a track with Technical Support. For information on other ways to contact Tech Support, refer to: http://support.sas.com/techsup/contact/index.html.

Thanks!

Jonathan