function [model,maxC] = kfoldCV(classification,trainingData,k,cExpMax,maxDataPerClass)
% kfoldCV does a k-fold Crossvalidation of parameter c for SVM
    noClasses=size(unique(classification),1);
    if cExpMax>0 %CV only if necessary
        cvAccuracy=zeros(2*cExpMax+1,1);
        for cExp=1:2*cExpMax+1 % try different values for c
            c=2^(cExp-cExpMax-1);

            randomMapping=transpose(randperm(size(trainingData,1)));
            accuracy=zeros(k,3);

            parfor i=1:k
                % split in training and test set
                trainData=trainingData(mod(randomMapping,k)+1~=i,:,:);
                testData=trainingData(mod(randomMapping,k)+1==i,:,:);
                trainClasses=classification(mod(randomMapping,k)+1~=i);
                testClasses=classification(mod(randomMapping,k)+1==i);
                %fprintf('i=%i, k=%i, c=%i\n',i,k,c)
                [trainClasses,trainData]=balanceClasses(trainClasses,trainData,...
                    maxDataPerClass,noClasses);
                model=svmtrain(trainClasses,trainData(:,:),sprintf('-t 0 -c %f -q',c));
                [~, accuracy(i,:), ~]=svmpredict(testClasses, testData(:,:), model,'-q');
            end
            %for each value for c take mean accuracy as accuracy
            cvAccuracy(cExp)=mean(accuracy(:,1));
        end
        [~,maxC]=max(cvAccuracy);
    else
        maxC=1; %no gridsearch since only one C possible
    end
    bestC=2^(maxC-cExpMax-1);
    [balancedClasses, balanceData]=balanceClasses(classification,trainingData,...
        maxDataPerClass,noClasses);
    model=svmtrain(balancedClasses, balanceData(:,:),sprintf('-t 0 -c %f -q',bestC));

end