If you you specify the Extreme Percentiles Default Limits Method in the SAS^® Enterprise Miner™ Replacement node, and if your data contains a variable with a frequency role, then the results of the node might be incorrect.

There are no errors or warnings to indicate that a problem occurred.

To work around the problem, do not use a frequency role. Instead, "expand" your data.

Example

Suppose that you want the Extreme Percentiles method, but your data set contains a variable called MY_FREQ_VARIABLE that you are using in a frequency role.

Delete the connection between your Replacement node and the predecessor node that contains the training data with frequency variable. Insert a SAS Code node in your path immediately before the Replacement node. Connect your predecessor node to the SAS Code node. Enter the following code using the Code Editor:

data &em_export_train;
   set &em_import_data;
   do my_temp_variable=1 to int(my_freq_variable);
      output;
      end;
   drop my_temp_variable my_freq_variable;
   run;

Close the SAS Code node, and run it. The code will expand the data by repeating each observation according to the value of the MY_FREQ_VARIABLE variable.

Connect the SAS Code node to your Replacement node, and run the Replacement node. The Replacement node will not use the variable MY_FREQ_VARIABLE as a frequency role because the SAS Code node dropped the variable. The Replacement node will use the correct number of observations because the SAS Code node created an expanded data set that does not need the frequency variable.

Operating System and Release Information