Matlab-使用Matlab实现的聚类算法-Clustering.zip

function [label, center, bCon, sumD, D] = litekmeans(X, k, varargin) %LITEKMEANS K-means clustering, accelerated by matlab matrix operations. % % label = LITEKMEANS(X, K) partitions the points in the N-by-P data matrix % X into K clusters. This partition minimizes the sum, over all % clusters, of the within-cluster sums of point-to-cluster-centroid % distances. Rows of X correspond to points, columns correspond to % variables. KMEANS returns an N-by-1 vector label containing the % cluster indices of each point. % % [label, center] = LITEKMEANS(X, K) returns the K cluster centroid % locations in the K-by-P matrix center. % % [label, center, bCon] = LITEKMEANS(X, K) returns the bool value bCon to % indicate whether the iteration is converged. % % [label, center, bCon, SUMD] = LITEKMEANS(X, K) returns the % within-cluster sums of point-to-centroid distances in the 1-by-K vector % sumD. % % [label, center, bCon, SUMD, D] = LITEKMEANS(X, K) returns % distances from each point to every centroid in the N-by-K matrix D. % % [ ... ] = LITEKMEANS(..., 'PARAM1',val1, 'PARAM2',val2, ...) specifies % optional parameter name/value pairs to control the iterative algorithm % used by KMEANS. Parameters are: % % 'Distance' - Distance measure, in P-dimensional space, that KMEANS % should minimize with respect to. Choices are: % {'sqEuclidean'} - Squared Euclidean distance (the default) % 'cosine' - One minus the cosine of the included angle % between points (treated as vectors). Each % row of X SHOULD be normalized to unit. If % the intial center matrix is provided, it % SHOULD also be normalized. % % 'Start' - Method used to choose initial cluster centroid positions, % sometimes known as "seeds". Choices are: % {'sample'} - Select K observations from X at random (the default) % 'cluster' - Perform preliminary clustering phase on random 10% % subsample of X. This preliminary phase is itself % initialized using 'sample'. An additional parameter % clusterMaxIter can be used to control the maximum % number of iterations in each preliminary clustering % problem. % matrix - A K-by-P matrix of starting locations; or a K-by-1 % indicate vector indicating which K points in X % should be used as the initial center. In this case, % you can pass in [] for K, and KMEANS infers K from % the first dimension of the matrix. % % 'MaxIter' - Maximum number of iterations allowed. Default is 100. % % 'Replicates' - Number of times to repeat the clustering, each with a % new set of initial centroids. Default is 1. If the % initial centroids are provided, the replicate will be % automatically set to be 1. % % 'clusterMaxIter' - Only useful when 'Start' is 'cluster'. Maximum number % of iterations of the preliminary clustering phase. % Default is 10. % % % Examples: % % fea = rand(500,10); % [label, center] = litekmeans(fea, 5, 'MaxIter', 50); % % fea = rand(500,10); % [label, center] = litekmeans(fea, 5, 'MaxIter', 50, 'Replicates', 10); % % fea = rand(500,10); % [label, center, bCon, sumD, D] = litekmeans(fea, 5, 'MaxIter', 50); % TSD = sum(sumD); % % fea = rand(500,10); % initcenter = rand(5,10); % [label, center] = litekmeans(fea, 5, 'MaxIter', 50, 'Start', initcenter); % % fea = rand(500,10); % idx=randperm(500); % [label, center] = litekmeans(fea, 5, 'MaxIter', 50, 'Start', idx(1:5)); % % % See also KMEANS % % [Cite] Deng Cai, "Litekmeans: the fastest matlab implementation of % kmeans," Available at: % http://www.zjucadcg.cn/dengcai/Data/Clustering.html, 2011. % % version 2.0 --December/2011 % version 1.0 --November/2011 % % Written by Deng Cai (dengcai AT gmail.com) if nargin < 2 error('litekmeans:TooFewInputs','At least two input arguments required.'); end [n, p] = size(X); pnames = { 'distance' 'start' 'maxiter' 'replicates' 'onlinephase' 'clustermaxiter'}; dflts = {'sqeuclidean' 'sample' [] [] 'off' [] }; [eid,errmsg,distance,start,maxit,reps,~,clustermaxit] = getargs(pnames, dflts, varargin{:}); if ~isempty(eid) error(sprintf('litekmeans:%s',eid),errmsg); end if ischar(distance) distNames = {'sqeuclidean','cosine'}; j = strcmpi(distance, distNames); j = find(j); if length(j) > 1 error('litekmeans:AmbiguousDistance', ... 'Ambiguous ''Distance'' parameter value: %s.', distance); elseif isempty(j) error('litekmeans:UnknownDistance', ... 'Unknown ''Distance'' parameter value: %s.', distance); end distance = distNames{j}; else error('litekmeans:InvalidDistance', ... 'The ''Distance'' parameter value must be a string.'); end center = []; if ischar(start) startNames = {'sample','cluster'}; j = find(strncmpi(start,startNames,length(start))); if length(j) > 1 error(message('litekmeans:AmbiguousStart', start)); elseif isempty(j) error(message('litekmeans:UnknownStart', start)); elseif isempty(k) error('litekmeans:MissingK', ... 'You must specify the number of clusters, K.'); end if j == 2 if floor(.1*n) < 5*k j = 1; end end start = startNames{j}; elseif isnumeric(start) if size(start,2) == p center = start; elseif (size(start,2) == 1 || size(start,1) == 1) center = X(start,:); else error('litekmeans:MisshapedStart', ... 'The ''Start'' matrix must have the same number of columns as X.'); end if isempty(k) k = size(center,1); elseif (k ~= size(center,1)) error('litekmeans:MisshapedStart', ... 'The ''Start'' matrix must have K rows.'); end start = 'numeric'; else error('litekmeans:InvalidStart', ... 'The ''Start'' parameter value must be a string or a numeric matrix or array.'); end % The maximum iteration number is default 100 if isempty(maxit) maxit = 100; end % The maximum iteration number for preliminary clustering phase on random % 10% subsamples is default 10 if isempty(clustermaxit) clustermaxit = 10; end % Assume one replicate if isempty(reps) || ~isempty(center) reps = 1; end if ~(isscalar(k) && isnumeric(k) && isreal(k) && k > 0 && (round(k)==k)) error('litekmeans:InvalidK', ... 'X must be a positive integer value.'); elseif n < k error('litekmeans:TooManyClusters', ... 'X must have more rows than the number of clusters.'); end bestlabel = []; sumD = zeros(1,k); bCon = false; for t=1:reps switch start case 'sample' center = X(randsample(n,k),:); case 'cluster' Xsubset = X(randsample(n,floor(.1*n)),:); [~, center] = litekmeans(Xsubset, k, varargin{:}, 'start','sample', 'replicates',1 ,'MaxIter',clustermaxit); case 'numeric' end last = 0;label=1; it=0; switch distance case 'sqeuclidean' while any(label ~= last) && it<maxit last = label; bb = full(sum(center.*center,2)'); ab = full(X*center'); D = bb(ones(1,n),:) - 2*ab; [val,label]