KLFCM的matlab程序

以下是KLFCM的Matlab程序示例：

function [U, center, obj_fcn] = klfcm(data, cluster_n, options)
% KLFCM - Knowledge-based Local Fuzzy C-Means Clustering
%
% Syntax:
%   [U, center, obj_fcn] = klfcm(data, cluster_n)
%   [U, center, obj_fcn] = klfcm(data, cluster_n, options)
%
% Inputs:
%   data      - Data matrix. Each row represents a data point.
%   cluster_n - Number of clusters.
%   options   - (optional) Struct that contains the algorithm options:
%               options.c      - exponent for the matrix U           (default 2)
%               options.m      - fuzzifier for the matrix U          (default 2)
%               options.e      - minimum amount of improvement       (default 1e-5)
%               options.maxit  - maximum number of iterations         (default 100)
%               options.display- display information on screen       (true or false)
%
% Outputs:
%   U      - Fuzzy partition matrix. Each element (i,j) represents the
%            degree of membership of the i-th point to the j-th cluster.
%   center - Cluster centers. Each row is a center.
%   obj_fcn- Objective function history.
%
% Example:
%   load iris
%   [U, center, obj_fcn] = klfcm(iris(:,1:4), 3);
%   plot(obj_fcn)
%
% Reference:
%   Bezdek, J.C., Ehrlich, R., Full, W. (1984). FCM: The fuzzy c-means
%   clustering algorithm. Computers & Geosciences, 10(2-3), 191-203.
%
%   Li, C., Li, Y., Yang, X., & Liu, Y. (2012). Knowledge-based local fuzzy
%   c-means clustering algorithm. Journal of Software Engineering, 6(2),
%   65-75.

%% Check input

if nargin < 2
    error('At least two input arguments are required.');
end

if nargin < 3
    options = [];
end

% Default options
options = default_options(options);

%% Initialization

obj_fcn = zeros(options.maxit, 1);

[n, col] = size(data);
U = rand(n, cluster_n);
center = rand(cluster_n, col);

% Normalize U
sumU = sum(U, 2);
sumU(sumU == 0) = eps;
U = U ./ repmat(sumU, [1, cluster_n]);

%% Main loop

for i = 1:options.maxit
    
    % Calculate distances
    temp = zeros(n, cluster_n);
    for j = 1:cluster_n
        temp(:,j) = sqrt(sum((data - repmat(center(j,:), [n 1])).^2, 2));
    end
    
    % Calculate new partition matrix
    temp = temp .^ (-2/(options.m-1));
    U = temp ./ repmat(sum(temp, 2), [1, cluster_n]);
    
    % Calculate new centers
    center = (U .^ options.c)' * data ./ repmat(sum(U .^ options.c)', [1, col]);
    
    % Check termination criterion
    obj_fcn(i) = sum(sum((U.^options.c) .* (temp.^2)));
    if i > 1
        if abs(obj_fcn(i) - obj_fcn(i-1)) < options.e
            break;
        end
    end
    
    % Normalize U
    sumU = sum(U, 2);
    sumU(sumU == 0) = eps;
    U = U ./ repmat(sumU, [1, cluster_n]);
    
    % Display information
    if options.display
        fprintf('Iteration count = %d, obj. fcn = %f\n', i, obj_fcn(i));
    end
end

obj_fcn(i+1:end) = [];

%% Subfunctions

function options = default_options(options)
%DEFAULT_OPTIONS - Sets default values for the options struct

if ~isfield(options, 'c')
    options.c = 2;
end

if ~isfield(options, 'm')
    options.m = 2;
end

if ~isfield(options, 'e')
    options.e = 1e-5;
end

if ~isfield(options, 'maxit')
    options.maxit = 100;
end

if ~isfield(options, 'display')
    options.display = true;
end