cc.glsn.v15.neuralnet

Class SigmoidFunction

java.lang.Object

cc.glsn.v15.neuralnet.SigmoidFunction

All Implemented Interfaces:

NetFunction, Serializable

public class SigmoidFunction
extends Object
implements NetFunction

Implementation of the sigmoid function for neural network activation. This is a common function to use for neural networks and recommended if you don't know which one to pick. Note: the sigmoid function can only return between 0 and 1. And near the edges of that, it takes an extreme input to get that. I may be doing something majorly wrong, but I'd recommend expecting outputs in the area of 0.25 - 0.75. If you are expecting boolean outputs, I'd do 0.25 = false, 0.75 = true or something like that. If you are expecting floating point output, I'd map the things you care about to 0.25-0.75 range. This seems to work well for me, but I'm no expert in neural networks or maths.

g(x)= 1/(1+e^(-x)) g'(x) = g(x) * (1 - g(x))

See Also:

Serialized Form