The NNET Procedure

ARCHITECTURE Statement

ARCHITECTURE architecture-option;

The ARCHITECTURE statement specifies the architecture of the neural network to be trained.

You can specify one of the following architecture-options (GLIM and MLP DIRECT are not allowed when an autoencoder is used; for more information about the autoencoder, see the section Autoencoder):

GLIM: specifies a neural network architecture that has no hidden layers (this is equivalent to a generalized linear model). If you specify this architecture-option, the HIDDEN statement is not allowed.
MLP: specifies a multilayer perceptron architecture that has one or more hidden layers. This is the default architecture. This option is the default for autoencoding.
MLP DIRECT: specifies that direct connections between each input and each target neuron be included when the MLP architecture is used.

When you use PROC NNET to train a neural network, this statement is optional. This statement is not allowed when you use PROC NNET to perform stand-alone scoring.

Note: If you specify the AUTOTUNE statement, PROC NNET uses the architecture suggested by the tuning optimization. As a result, it might use a different architecture from the one specified in the ARCHITECTURE statement.

Last updated: November 11, 2020