The TARGET statement specifies the target variable for the neural network. If you do not specify a TARGET statement, PROC NNET trains an autoencoder. For more information, see the section Autoencoder.
You can specify the following options:
-
ACT=EXP |IDENTITY |SIN |SOFTMAX |TANH
specifies the activation function for the target.
You can specify the following values:
- EXP
specifies the exponential function. You can use ACT=EXP only with ERROR=GAMMA or ERROR=POISSON.
- IDENTITY
specifies the identity function.
- SIN
specifies the sine function.
- SOFTMAX
specifies the softmax function.
- TANH
specifies the hyperbolic tangent function.
For the GLIM architecture, you can only specify ACT=IDENTITY for an interval target and ACT=SOFTMAX for a nominal target, which are the same by default. For the MLP or MLP DIRECT architecture, the SOFTMAX method is used only with a nominal target, whereas the other methods are used only with an interval target. By default, ACT=IDENTITY for the interval target and ACT=SOFTMAX for the nominal target.
-
COMB=ADD |LINEAR
specifies the combination function for the target layer. You can specify the following combination functions:
- ADD
specifies the additive combination function.
- LINEAR
specifies the linear combination function.
By default, COMB=LINEAR.
-
ERROR=ENTROPY |GAMMA |NORMAL |POISSON
specifies the error function. The entropy error function is used only when LEVEL=NOM. You can specify the following error functions:
- ENTROPY
specifies the cross-entropy function.
- GAMMA
specifies the gamma error function. This function is usually used when you want to predict the time between events. Only ACT=EXP is valid when ERROR=GAMMA.
- NORMAL
specifies the normal error function, which is the sum of the squared differences between the network output and the target value.
- POISSON
specifies the Poisson error function. This function is usually used when you want to predict the number of events per unit time. Only ACT=EXP is valid when ERROR=POISSON.
By default, ERROR=NORMAL when LEVEL=INT, and ERROR=ENTROPY when LEVEL=NOM.
-
LEVEL=INTERVAL |NOMINAL
specifies the variable type. You can specify the following values:
- INTERVAL
specifies that the variable is interval, which must be numeric.
- NOMINAL
specifies that the variable is nominal, also known as a classification variable, which can be numeric or character.
By default, LEVEL=INTERVAL.