classification

I want to learn how to do classification using Neural Network in Matlab Having Elliptical Basis Function , Can any one help .... Code is attached

What I know is that first I need to create NN and then make Decision Making

Function ( Elliptical Basis Function )

 tic 

    maxround = 5;
    hiddenLayerSize = [3 5 7 10 13];

    errors = zeros(maxround,1);
    trainPerformance = zeros(maxround,1);
    valPerformance = zeros(maxround,1);
    testPerformance = zeros(maxround,1);
    timedata = zeros(maxround,1);
    NoofNeurons = zeros(maxround,1);
    Accuracy = zeros(maxround,1);
    TestFold = zeros(maxround,1); 
  NoOfClasses = zeros(maxround,1); 
 NoofInstances = zeros(maxround,1); 
 SizeofInputLayer = zeros(maxround,1); 

 for i=1: maxround

    net = patternnet(hiddenLayerSize);

    net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};
    NoOfinputs  =   net.inputs

    net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'};
    NoOfOutPuts  =   net.outputs

    net.divideFcn = 'dividerand';  % Divide data randomly
    net.divideMode = 'sample';  % Divide up every sample
    net.divideParam.trainRatio = 70/100;
    net.divideParam.valRatio = 15/100;
    net.divideParam.testRatio = 15/100;
    net.trainFcn = 'trainlm';  % Levenberg-Marquardt
    net.performFcn = 'mse';  % Mean squared error
    [net,tr] = train(net,inputs,targets);
    outputs = net(inputs);

    errors = gsubtract(targets,outputs);
    performance(i) = perform(net,targets,outputs);
    trainTargets = targets .* tr.trainMask{1};
    valTargets = targets  .* tr.valMask{1};
    testTargets = targets  .* tr.testMask{1};
    trainPerformance(i) = perform(net,trainTargets,outputs);
    valPerformance(i) = perform(net,valTargets,outputs);
    testPerformance(i) = perform(net,testTargets,outputs);

    NoofNeurons(i) = hiddenLayerSize(1);
    NameofDataSet  = 'Heart';
    TestFold(i) = i;
    NoOfClasses(i) =  size(targets,1); % Number of classed to be classified 
    NoofInstances(i) = size(targets,2); % Number of Instances 
    SizeofInputLayer(i) = size(inputs,2);

 end

[Walter Roberson]

Thank you for taking the time to format your code!

Answer 1

[Greg Heath]

http://groups.google.com/group/comp.soft-sys.matlab/msg/1e4cc0229688dd7a

http://groups.google.com/group/comp.soft-sys.matlab/msg/f802608bf7d07df0

Hope these help.

Greg

[Greg Heath]

IGNORE THIS COMMENT OF 8 JAN 2012. IT IS BASED ON FAULTY LOGIC. FOR ANY OTHER VALUE OF THE LINEAR COMBINATION THAN ZERO, THE CONTOURS ARE ELLIPTIC. =========================================================================== WHOOPS! MY APOLOGIES!

I have just realized that , in general, NEWFF with RADBAS will not create EBFs. Consider a two-input net with w11*x1+w12*x2 + b1 the argument of RADBAS.

Obviously, exp(-(w11*x1+w12*x2+b1)^2) is equal to 1 along the line 0 =w11*x1+w12*x2+b1. Therefore the constant level contours are not ellipses.

Sorry for the bum steer.

Greg

[Puneet Arora]

Then what do you suggest .. I need to change the H and O , My I-H-O is structure : 1 (inputs)-H- 4 (Outputs) and H = [1 2 4 8 16 32 64]

net = newff(minmax(ptrn),[H O],{'radbas', 'purelin'});

I m doing classification iris dataset ...