The HPNEURAL Procedure

PARTITION Statement

PARTITION ROLEVAR=variable( TRAIN=number | VALIDATE=number );

PARTITION FRACTION( TRAIN=number | VALIDATE=number );

The PARTITION statement specifies how to divide the input data set into a training subset and a validation subset.

The statement implements two alternate methods of specifying the split between the training and validation data. Either you can explicitly specify training observations and validation observations by specifying ROLEVAR=variable, where variable is a variable in the input data set, or you can specify that an approximate fraction of the input data set be used for training observations or validation observations by specifying FRACTION( TRAIN=number ) or FRACTION( VALIDATE=number ).

ROLEVAR=variable( TRAIN=number | VALIDATE=number ): specifies that the variable in the input data set be used to decide whether an observation is used for training or for validation. You can either specify the value used to identify training observations or the value used to identify validation observations. If you specify TRAIN=number, then an observation is used for training if the value of variable equals number; otherwise the observation is used for validation. If you specify VALIDATE=number, then an observation is used for validation if the value of variable equals number; otherwise the observation is used for training.
FRACTION( TRAIN=number | VALIDATE=number ): specifies the approximate fraction of the input data set to be used for training or validation. If you specify TRAIN=number, then approximately the fraction of the data set specified by number is used as training observations, and the rest are used for validation observations. If you specify VALIDATE=number, then approximately the fraction of the data set specified by number is used as validation observations. The split between training and validation observations can only approximate the requested fraction because that fraction is used as a cutoff value for a random number generator to determine the actual split. If you require a more accurate split, you must use the ROLEVAR= option to specify the split explicitly.

When you are training, the PARTITION statement is optional. If you do not include the PARTITION statement, every fourth observation (starting with the first observation) is used as a validation observation, unless you specify VALID=NONE in the TRAIN statement. In this case, no validation is performed. The PARTITION statement is not allowed when you are doing stand-alone scoring.

Fit statistics reported after training are only computed using validation observations. Fit statistics reported after stand-alone scoring are computed using all observations.

Last updated: May 25, 2022