function [predictedValue] = myNeuralNet(testSet, trainSet)

% design a regression neural network based on radial basis networks
% note how the matrices are transposed
% independent variable matrix is an RxQ matrix, where Q is instance number
% dependent variable is a matrix SxQ, where Q is instance number
spreads = [10:10:50]';
bestSpread = 1;

% find a good spread value by using validation set
validationSet = trainSet;
minErr = inf;
for i = 1:size(spreads,1)
    validationInstance = validationSet(i,:);
    validationSet(i,:) = [];
    myTempNet = newgrnn(validationSet(:,1:(size(validationSet,2)-1))', trainSet(:,size(validationSet,2))',spreads(i));
    tempPrediction = sim(myTempNet,validationInstance(1,1:(size(validationInstance,2)-1))')
    tempError = abs(validationInstance(1,size(validationInstance,2)) - tempPrediction)/validationInstance(1,size(validationInstance,2));
    if tempError < minErr
        minErr = tempErr;
        bestSpread = spreads(i);
    end
end
% simulate network for a new input
% i.e. evaluate it for the test instance 
% then store the predicted value
myNet = newgrnn(trainSet(:,1:(size(trainSet,2)-1))', trainSet(:,size(trainSet,2))',spreads(i));
predictedValue = sim(myNet,testSet(1,1:(size(trainSet,2)-1))');

end