In SAS^® Enterprise Miner™, bagged models are created by iteratively modeling re-sampled observations. The model from each iteration contributes to the final total score. Some of the lines in the Flow code of the modeling node assign values from the first iteration. However, those lines are incorrectly missing from the Score code so that the model from the first iteration does not contribute to the total score.

There are no error or warning messages to indicate a problem. The only way to tell that the problem occurred is to inspect the created Score code and compare it to the Flow code. There is no workaround.

Example: suppose that your target variable is called TARGET and that it takes the values "yes" and "no." The following lines appear just past the end of the first iteration in the Flow code:

*--------------------------------------------------------------*;
* 
Bagging: Saving Probabilities for Next Iteration;
 
*---------------------------------------------------------------*; 
EndGrp_P_TARGETyes = P_TARGETyes;
EndGrp_P_TARGETno  = P_TARGETno;

These lines should also appear in the Score code. However, they do not. These lines assign the predicted probability from the first iteration model to variables of the form EndGrp_P_<varname><varlevel>. Following each iteration, the code creates a weighted average of the value of this Endgrp_P_ variable with average predicted probabilities from latter iterations to create the predicted probability from bagging.

Operating System and Release Information