Outputs not within 0-1 though SigmoidActivationFunction for ElmanPattern?

published by DennisN on Wed, 06/20/2012 - 01:49

I created an ElmanPattern with an ActivationFunctionSigmoid and have recieved
output values as -0.7 and 1.5.

Started to look in the source code for ElmanPattern in Encog 3.1.0 and found that the output layer is created with
network.addLayer(new BasicLayer(null, false, this.outputNeurons));
though the API documentation says 'The specified activation function will be used on all layers.'
I changed it to
network.addLayer(new BasicLayer(this.activation, true, this.outputNeurons));
and seems to work fine.

Don't know which way it's supposed to be ?

Neural Network Forums:

Using Encog in Java