Simply Salford Blog

How to Run a Model Using GPS

Posted by David Tolliver on Wed, Aug 27, 2014 @ 05:30 AM

We recently had a question about running a model using GPS, and wanted to share the answer in case anyone else has the same issue.

Say you have 16K records and 5k variables, 0/1 Flag for the Dependent variable. You may run TreeNet and identify 360 important variables, and want to keep those variables and run a model using GPS.                                                                                                  How would you accomplish this task using SPM?

There are a few different ways to do this depending on what version you have. 

The first is available in all versions. 
1 )  Once you have grown your TreeNet model, click the [Summary...] button found on the TreeNet output window.
2.)  From the resulting Summary window, select the Variable Importance tab.
3.)  From within the Variable Importance select your 360 variable and click [New Keep List] button  (example below).

DataMining.TreeNet1

4.) This results in a SPM Notepad with a new KEEP list of the selected 360 variables (example below with much fewer variables).

DataMining.TreeNet2

5.)  With the SPM Notepad in the foreground, select Submit Window from the File menu.  The result is that SPM is now prepared to model using only the 360 variables selected and represented in the KEEP statement displayed in the SPM Notepad.  You are now ready to run a GPS.

DataMining.TreeNet3

6.) Open the Model Setup dialog by clicking the Model Setup toolbar icon, or selecting Construct Model... from the Model menu.  Note that the Model menu is only available when the Classic Output window is in the foreground.

                                                           describe the image  or   describe the image

7.)  From the Model Setup dialog, select GPS as the Analysis Method and you are ready to set up your model and run.


DataMining.TreeNet6

 

 

 

how to build a predictive model

Topics: SPM, TreeNet, data mining, GPS, predictive model, data analysis, Tutorial, model scoring