In SAS® Enterprise Miner™ some nodes produce splits of the data. Examples: Interactive Binning, Gradient Boosting, Interactive Grouping, Decision Tree. When you view the results of nodes such as these, sometimes the following symptoms occur.
NOTE: Will not search for split on variable variable name.
In practice, these messages are displayed when some of the samples have an insufficient number of events and non-events. The underlying procedure might not be able to iterate. Sampling is used at different points to determine split values, and then the model is fit to the whole data set. If there are not enough events, then SAS Enterprise Miner cannot determine where the splits should occur.
These messages might occur for the above nodes even when other modeling nodes run.
Example: a Regression node completes successfully, but the Gradient Boosting node does not find splits. In this example, the results of the Regression node usually show that there are many near-zero standard errors for the parameter estimates. Those values suggest that there is not enough variability in the covariates with respect to the number of events and non-events. Sometimes infinite standard errors occur. See SAS Note 22599: "Understanding and correcting complete or quasi-complete separation problems" for additional details.
Beginning in SAS Enterprise Miner 12.3, additional properties are available that enable you to change the size of the sample. Try increasing the size of the sample, and test whether the message is resolved. Some trial and error might be involved with selecting a large-enough sample. You might need to specify a sample size that is as large as the number of observations in the data.