REGULARIZATION Subcommand (CATREG command)

REGULARIZATION specifies the method for regularized regression.

The specification on REGULARIZATION is keyword NONE, RIDGE, LASSO or ENET. If REGULARIZATION is not specified, NONE is the default. Also, a dataset name or filename must be specified. The statistics and coefficients and, if applicable, the estimated prediction error, for all regularized models will be written to this file.

NONE. No regularization.

RIDGE(start value, stop value, increment)('savfile'|'dataset'). Ridge Regression. A value list in parentheses following the keyword should be given. The first value specifies the start value of the penalty parameter, the second value the stop value, and the third value specifies the increment.

LASSO(start value, stop value, increment)('savfile'|'dataset'). LASSO (Least Absolute Shrinkage and Selection Operator). A value list in parentheses following the keyword should be given. The first value specifies the start value of the penalty parameter, the second value the stop value, and the third value specifies the increment.

ENET(start, stop , incr)(start, stop , incr)('savfile'|'dataset'). Elastic Net. Two value lists in parentheses following the keyword should be given. The first list specifies the start, stop, and increment values for the Ridge penalty, the second list specifies the start, stop, and increment values for the Lasso penalty.

  • If a REGULARIZATION method is specified without specification of a resample method at the RESAMPLE subcommand or specification of test cases at the SUPPLEMENTARY subcommand, any other keywords than REGU at the PRINT and PLOT subcommands are ignored. Also, the SAVE subcommand, and the TRDATA keyword at the OUTFILE subcommand are ignored.
  • If a a resample method is specified at the RESAMPLE subcommand, or if test cases are specified at the SUPPLEMENTARY subcommand, specified PRINT, PLOT, SAVE, and OUTFILE output will be given for the model with lowest prediction error or with lowest Test MSE.
  • Output of an analysis with a specific value of the penalty parameters is obtained by setting the start values to specific penalty values, the stop values equal to the start values and the increment values to 0.