The LOGSELECT Procedure

The functionality of the LOGSELECT procedure closely resembles that of the HPLOGISTIC procedure, which is a high-performance procedure. The LOGSELECT procedure is the next generation of the HPLOGISTIC procedure, and it was developed specifically for SAS Viya. Both procedures are designed to run on a cluster of machines that distribute the data and the computations. Both procedures perform computations in multiple threads.

Both the LOGSELECT and HPLOGISTIC procedures fit and perform model selection for logistic regression models. The models can contain main effects that consist of both continuous and classification variables and interaction effects of these variables. The HPLOGISTIC procedure provides GLM and reference parameterizations of classification variables; the LOGSELECT procedure provides the full set of parameterizations that are available in PROC LOGISTIC and other modeling procedures. The LOGSELECT procedure is additionally capable of creating complex constructed effects, including univariate spline and polynomial expansions.

With the LOGSELECT and HPLOGISTIC procedures, you request model selection by using the SELECTION statement. Both procedures offer the same methods of effect selection, but the LOGSELECT procedure also provides LASSO selection and produces selection plots by using ODS Graphics.

The LOGSELECT procedure provides more regression diagnostics than the HPLOGISTIC procedure. This release of the LOGSELECT procedure is more limited in postfitting functionality than the HPLOGISTIC procedure.

The LOGSELECT procedure provides the full set of parameterizations that are available in PROC LOGISTIC. The LOGSELECT procedure uses the GLM parameterization for the CLASS variables by default. The LOGISTIC procedure uses the EFFECT parameterization for the CLASS variables by default. In either procedure, you can use the PARAM= option in the CLASS statement to change the parameterization.

The LOGISTIC procedure uses Fisher scoring by default. The LOGSELECT procedure uses a modification of the Newton-Raphson algorithm with a ridged Hessian by default. You can choose different optimization techniques, including first-order methods that do not require a crossproducts matrix or Hessian, by using the TECHNIQUE= option in the PROC LOGSELECT statement.

The LOGSELECT procedure provides a richer set of model selection options than the LOGISTIC procedure, including LASSO selection, information-criterion-based selection and stopping criteria, and validation-based criteria.

The LOGSELECT procedure uses ODS Graphics to display plots that help interpret the selection process; the LOGISTIC procedure displays many diagnostic and fit plots.

The LOGISTIC procedure offers a wide variety of postfitting analyses, such as contrasts, estimates, tests of model effects, least squares means, and odds ratios. This release of the LOGSELECT procedure is limited in postfitting functionality, because for large data sets the focus is primarily on model fitting and scoring.

The LOGSELECT procedure is specifically designed to operate in SAS Viya and performs computations in multiple threads. The LOGISTIC procedure executes in a single thread on a single machine.