kmeans和IOSDATA算法matlab程序

function [centroid, result] = ISODATA(data, iteration, desired_k, minimum_n, maximum_variance, minimum_d) % also same to traditional kmeans centroid = data(randperm(size(data,1),desired_k)', :); distance_matrix = zeros(size(data,1), desired_k); result = zeros(size(data,1),1); for i = 1 : iteration previous_centroid = centroid; for j = 1 : size(distance_matrix,1) for k = 1 : size(distance_matrix,2) distance_matrix(j,k) = sqrt(sum((data(j,:)-centroid(k,:)) .^ 2)); end end [~,result] = min(distance_matrix,[],2); % decide whether the number of samples in each class is smaller than minimum_n for j = 1 : size(centroid, 1) if(isempty(find(result == j,1)) || (size(find(result == j),1) < minimum_n)) % one class with number of samples in it less than minimum_n % should be deleted, along with records in the distance_matrix centroid(j,:) = []; distance_matrix(:,j) = []; % reassign each sample to its closet neighbor class [~,result] = min(distance_matrix,[],2); % recalculate centroid locations for k = 1:size(centroid,1) centroid(k,:) = mean(data(result(:,1) == k,:)); end end end % decide if combine and split operations are needed % 1st: too few classes if(size(centroid,1) <= (desired_k/2)) [centroid] = ISODATA_split(data, centroid, result, minimum_n, maximum_variance); % 2nd: too many classes elseif(size(centroid,1) > (2*desired_k)) [centroid] = ISODATA_combine(centroid, result, minimum_d); end if(previous_centroid == centroid) fprintf('Clustering over after %i iterations...\n',i); break; end end end