NNET Procedure

The NNET procedure trains a multilayer perceptron neural network in SAS Viya Workbench. For more information about multilayer perceptron neural networks, see Bishop (1995). PROC NNET can also use a previously trained network to score a data table (referred to as stand-alone scoring), or it can generate SAS DATA step statements that can be used to score a data table.

Training a multilayer perceptron neural network requires the unconstrained minimization of a nonlinear objective function. Because there are currently no practical methods to guarantee finding a global minimum of that objective function, one way to be reasonably sure of finding a good solution is to train the network multiple times by using different sets of initial values for the weights.