simple k-means algorithm implemented in GNU Octave

kMeans.m

function y = kMeans(m, k, i)
% m: input matrix 
%   row:    each column is a sample
%   column: sample number
% k: k clusters
% y: vector consist of the cluster number of each sample

[sampleNum, featureNum] = size(m);
centers = m(1:k, :);
y = zeros(sampleNum,1);
iterNum = 10000000;
cuttentIter = 0;
precision = 1e-8;

% endless loop
while 1

% calculate distance between every sample and every cluster center
distMatrix = zeros(sampleNum, k);
for i = 1:sampleNum
    for j = 1:k
        distMatrix(i,j) = norm(m(i,:)-centers(j,:), 2);
    end
end

% update cluster according to distance
for i = 1:sampleNum
    [~, y(i)] = min(distMatrix(i,:));
end

% update cluster centers
nSamplesInACluster = zeros(k,1);
lastCenters = centers;
centers = zeros(size(centers));
for i = 1:sampleNum
    currentCluster = y(i);
    centers(currentCluster,:) = centers(currentCluster,:).*nSamplesInACluster(currentCluster) + m(i,:);
    nSamplesInACluster(currentCluster)=nSamplesInACluster(currentCluster)+1;
    centers(currentCluster,:) = centers(currentCluster,:)./nSamplesInACluster(currentCluster);
end

% finish iteration condition
if(cuttentIter >= iterNum || max(abs(centers - lastCenters)) < precision )
    break;
end

end % while


end % function kMeans