Rejecting high percent mode in Enterprise Miner

[TrishPaquette]

In the case study solution to Module 3, the author rejects variables with a high percent mode (>97%). Is this done in the metadata node? I can choose statistics to filter on, but mode is not an option. 

[RobertBlanchard]

Hello Trish,

Yes, the variables with the large mode percentage are set to rejected using the metadata node.  

One alternative to the provided solution may be to set the data partition node to perform a stratified simple random sample, stratifying on those variables with a large mode to ensure all values in the variables exist in each partition.  To do this, click the variables ellipsis in the data partition node.  Then change the Partition Role value from default to  Stratification for the categorical variables of interest.  If a level is VERY rare, then sometimes this approach may not work and you'll receive an error from the node.

Good luck!

Best,

  Robert 

[TrishPaquette]

Revisiting this because I can't seem to reject automatically in
the Metadata node. I am offered some statistics on which to reject, but
percent mode is not one of them. How can I do this automatically?

[RobertBlanchard]

Hey,

I believe one would have to reject the variables manually in this scenario.

Best,

  Robert