Neural Network Normalization process

Hello all,

I have a question regarding the NN normalization procedure. When a NN is trained using the train(net,x,y) command the function somehow normalizes x and y in order to ensure that the weights and biases of the network are bound by [-1,1].

Currently I am in the process of trying to apply a set of NN weights and biases analytically (instead of just calling net(xtest)) using the following equation:

ytest = Outputbias+Hiddenweight*tanh(Inputbias + Inputweight*xtest);

which produces an output -- however it does not produce the same output as (ytest = net(xtest)).

I'm assuming the difference is due to the fact that xtest is not normalized before using the above equation.

I tried to simply divide xtest by its maximum before feeding it in to the equation however the results still differ.

Does anyone know how xtest should be manipulated in order to produce the same output as net(xtest).

Thanks! Bryan

Answer 1

[Greg Heath]

http://www.mathworks.com/matlabcentral/newsreader/view_thread/341631#936181

Hope this helps.

Thank you for formally accepting my answer

Greg