Page 1 of 1

Why is solution not exact when "fixed"

PostPosted: Tue May 24, 2016 8:27 am
by Lawrence
FixedData.PNG
Solution and data - not the same
FixedData.PNG (137.23 KiB) Viewed 5526 times


Hi,

I am using

ActivationNetwork network = new ActivationNetwork(
new BipolarSigmoidFunction(sigmoidAlphaValue),
inputList.Count,
inputList.Count,
1);

BackPropagationLearning teacher = new BackPropagationLearning(network);

With the same data for input AND output - just to see process of weights being set.

So inputList.Count == 1;

Why does the solution line, not match the data line exactly. They are the same data...

Re: Why is solution not exact when "fixed"

PostPosted: Tue May 24, 2016 9:01 am
by andrew.kirillov
Hello,

Network's weights are set randomly initially. Then you teach it to get result you need. Obviously it may take certain number of iterations for weights to converge to "optimal". Play with number of iterations, learning parameters ...

Re: Why is solution not exact when "fixed"

PostPosted: Tue May 24, 2016 10:02 am
by Lawrence
Fixed2.PNG
Fixed2.PNG (124.22 KiB) Viewed 5523 times


Hi Andrew,

I have actually tried all manner of variants of the parameters but I still never get to an exact match - is that because of the Sigmoid?

I can't get better match than the picture above.