The HPNEURAL Procedure

Syntax: HPNEURAL Procedure

The following statements are available in the HPNEURAL procedure:

  • PROC HPNEURAL <DATA=SAS-data-set>  <NOPRINT> ;

  • PERFORMANCE performance-options;

  • ARCHITECTURE architecture-options;

  • ID variables;

  • INPUT variables </ <LEVEL=INT | LEVEL=NOM >
    <STD=RANGE | STD=ZSCORE | STD=NONE >
    <MISSING=MAP>>
    ;

  • WEIGHT variable | _INVERSE_PRIORS_;

  • HIDDEN number </ ACT= activation-function>;

  • TARGET variables </ <LEVEL=INT | LEVEL=NOM >
    <STD=RANGE | STD=ZSCORE | STD=NONE >
    <ACT=activation-function> <ERROR=error-function>>
    ;

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

  • PARTITION FRACTIONTRAIN=number | VALIDATE=number );

  • TRAIN <NUMTRIES=number> <MAXITER=number>
    <VALID=NONE> <OUTMODEL=SAS-data-set>
    ;

  • SCORE OUT=SAS-data-set <MODEL=SAS-data-set> ;

  • CODE FILE=’external-file’ | fileref;

When you train a neural network, the PROC HPNEURAL, INPUT, TARGET, and TRAIN statements are required. The HIDDEN statement is required unless you use the logistic architecture (in which case, the HIDDEN statement is not allowed).

When you use a previously trained neural network to score a data set, only the PROC HPNEURAL, SCORE, ID, PERFORMANCE, and CODE statements are allowed.

Last updated: May 25, 2022