function [TestPCAOutput, distMatrix] = computekNNClassificationPerformance(ftestPCA_all, TestPCALabels, sizeTest, stepPrint, numCoresKnn, param)

%Distance matrix

if strcmp(param.matchDistance, 'chisq')

    distMatrix = sc_pdist2(ftestPCA_all, ftestPCA_all, 'chisq');

elseif strcmp(param.matchDistance, 'euclidean')

    %fast Euclidean distance

    distMatrix = full(fastEuclideanDistance(ftestPCA_all, ftestPCA_all));

else %if strcmp

    distMatrix = pdist2(ftestPCA_all', ftestPCA_all', param.matchDistance);

end %if strcmp

%modo parallelo per knn (k = 1) classification

%leave-one-out

%we use knn_neighbors + 1 because otherwise it would find the same vector

%in this way we choose the second neighbor (which is the actual first neighbor)

%(we use the entire feature vector for all samples)

%loop on test samples

%init

TestPCAOutput = zeros(sizeTest, 1);

%parallel vars init

% ftestPCA_all = parallel.pool.Constant(ftestPCA_all);

knnDistancePar = param.knnDistance;

matchDistance = param.matchDistance;

numkPar = param.knn_neighbors;

% start_pool(numCoresKnn);

% parfor g = 1 : sizeTest

for g = 1 : sizeTest

    %get id of current worker

    t = getCurrentTask();

    %display progress

    if mod(g, stepPrint) == 0

        %fprintf(1, ['\t\tCore ' num2str(t.ID) ': ' num2str(g) ' / ' num2str(sizeTest) '\n'])

        fprintf(1, ['\t\t' num2str(g) ' / ' num2str(sizeTest) '\n'])

    end %if mod(i, 100) == 0

    if strcmp(knnDistancePar, matchDistance)

        %we can re-use the distance matrix

        distV = distMatrix(g, :);

        sortV = sort(distV, 'ascend');

        minD = sortV(2); %the first will be 0

        idx = find(distV == minD);

        idx = idx(1); %se dovessero essercene altri a pari merito  

    else %if strcmp(param.knnDistance, param.matchDistance)

        %we use + 1

        %idx = knnsearch(ftestPCA_all.Value', ftestPCA_all.Value(:, g)', 'K', numkPar + 1, 'Distance', knnDistancePar);

        idx = knnsearch(ftestPCA_all', ftestPCA_all(:, g)', 'K', numkPar + 1, 'Distance', knnDistancePar);

        idx(idx == g)  = []; %il pių vicino č il vettore stesso, lo togliamo

        idx = idx(1); %se dovessero essercene altri a pari merito

    end %if strcmp(param.knnDistance, param.matchDistance)

    TestPCAOutput(g) = idx;

end %for g

%mettiamo le labels al posto degli indici trovati

for g = 1 : sizeTest

    TestPCAOutput(g) = TestPCALabels(TestPCAOutput(g));

end %for g