You are here

Home » Neural Network Forums » Neural Networks » General Neural Networks

have a problem with activation function.

tamperen's picture
published by tamperen on Thu, 08/02/2012 - 16:56

my out put is like 1.2 , 1.4 .1 7 , 0.8 , 0. 5
none of them are negative

the issue i when i apply the sigmoid function it turn 1.2 = 1 , 1.4 =1
hence messing up my error variable

is there a way i can extend the range of the sigmoid function from 0 ~ n
n = any positive integer

i thought just multiplying it by 2 but it dont seems to work ,

and is there any better way to solve this?

Neural Network Forums: 
General Neural Networks
jeffheaton's picture

Submitted by jeffheaton on Wed, 08/08/2012 - 08:10.

You probably should normalize to the range (0,1) or (-1,1). You can create a new activation function, with a larger range. You would have to adjust the sigmoid or tanh function and compute the derivative. It would be a little(but not much) more complex than multiplying the sigmoid function by two.

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer