Optimization loop for number of neurons in neural network

Dear all,

I am trying to optimize the number of neurons in one-layer network in a simple "for"-loop. However, when I find the optimal number and try to establish the network again, I get a completely different result: the error is different. Knowing that the randomness is present, I try to fix the weights' assignment but it does not help.

I am definitely not an expert in ANN therefore probably can't simply understand where the problem comes from.

Here is my code

load data % loads X and Y tables

setdemorandstream(391418381)

SRmax = 0;

numNeuronsopt=0;

for i=1:20

numNeurons = 9+i;

net = patternnet(numNeurons);

[net,tr] = train(net,X,Y);

nntraintool

Xtest = X(:,tr.testInd);

Ytest = Y(:,tr.testInd);

Ytesthat = net(Xtest);

[MCR,ConfMat] = confusion(Ytest,Ytesthat);

fprintf('Percentage Correct Classification : %f%%\n', 100*(1-MCR));

fprintf('Percentage Incorrect Classification : %f%%\n', 100*MCR);

SR_temp = (1-MCR)*100;

if SR_temp>SRmax

    SRmax = SR_temp;

    numNeuronsopt = numNeurons;

end

end

Thank you for your help!