Java Problem With Double in Neural Network Trainig

See more:

Java

Hi guys,

I create the neural network for xor operation with guidance of Jeff Heaton's book "Introduction to Neural Network in Java".I am sure about there is no mistake in coding because it is true for books samples.Also it changes weights correctly when I gave same input(for example only input=(0,1) output=1) for thousands times. But when I give more than one input(all 4 possibilities for xor operation) for thousands of cycles the outputs are very close to each other.

My output for 100,000 cyles of training for each input;

input      actual output           ideal output
------     ------------------      -------------
0 , 0  => 0.003772846137554265           0
1 , 0  => 0.6224007788208237             1
0 , 1  => 0.6224007779452603             1
1 , 1  => 0.6224593302953872             0

Does anybody know why the output numbers are closing each other (it goes around 0,62) but it is true for first input pair? Can you help me?