Cross Validation in Neural Network ?

i want to do a cross validation 10% of the data should be for training and the rest is for test .

i dont know how to perform that.

filename='FIFA7.xlsx';
A =xlsread(filename);
[m,n]=size(A);
T= A(:,1);
data= A(:,(2:end));
rows80=int32(floor(0.8 * m));
trainingset=A(1:rows80,:);
testset=A(rows80+1:end,:);
t=trainingset(1:rows80,1);
t_test=A(rows80+1:end,1);
  
%    k=10
%    cvFols=crossvalind('kfold',l,k);
%    for i =1:k
%        testIdx=(cvFolds==i);
%        trainIdx=~test;
%    
%        
net= newff(trainingset',t');
y=sim(net,trainingset');
%net.trainParam.epoch=20;
net= train(net,trainingset',t');
y=sim(net,trainingset');
y_test=sim(net,testset');
 p=0;
  y1=hardlim(y');
   y2= hardlims(y_test);
  for(i=1:size(t,1))
      if(t(i,:)==y1(i,:))
     p=p+1;
      end
  end
  
  
 trainerror =100*p/size(trainingset,1);
  e=0;
  y2=hardlim(y_test');
 for(j=1:size(t_test,1))
    if(t_test(j,:)==y2(j,:))
         e=e+1;
   end
 end
 
   testerror=100*e/size(t_test,1)

Answer 1

[Greg Heath]

You may be confused. The MATLAB DEFAULT is RANDOM DATA DIVISION with

80% training

10% validation (tends to prevent overtraining)

10% testing

Depending om the size of the data, you can obtain 10 to 20 separate designs and choose the best.

Instead of storing all of the weights, I just store the current best net as I loop through the designs.

I am sure that I have posted zillions of examples in both the NEWSGROUP comp.soft.sys.matlab and ANSWERS.

Hope this helps.

Thank you for formally accepting my answer

Greg