www.pudn.com > Classification_toolbox.part01.rar > Competitive_learning.m

function [patterns, targets, label, W] = Competitive_learning(train_patterns, train_targets, params, plot_on) 
 
% Perform preprocessing using a competitive learning network 
% Inputs: 
% 	patterns	- Train patterns 
%	targets	    - Train targets 
%	params	    - [Number of partitions, learning rate] 
%   plot_on     - Plot while performing processing? 
% 
% Outputs 
%	patterns		- New patterns 
%	targets			- New targets 
%	label			- The labels given for each of the original patterns 
%   W               - Weights matrice 
 
max_iter       = 1000; 
[c, r]		   = size(train_patterns); 
[N, eta]	   = process_params(params); 
decay          = 0.99; 
 
%Preprocessing: 
% x_i <- {x_i, 1} 
x              = [train_patterns ; ones(1,r)]; 
%x_i <- x_i./||x_i|| 
x              = x ./ (ones(c+1,1) * sqrt(sum(x.^2))); 
 
%Initialize the W's 
i              = randperm(r); 
W              = x(:,i(1:N)); 
 
for i = 1:max_iter, 
    %Randomally order the patterns 
    order = randperm(r); 
    change= 0; 
     
    for k = 1:r, 
        J = W'*x(:,order(k)); 
        j = find(J == max(J)); 
         
        old_W   = W(:,j); 
         
        %W_j <- W_j + eta*x 
        W(:,j)  = W(:,j) + eta*x(:,order(k)); 
         
        %W_j <- W_j/||W_j|| 
        W(:,j)  = W(:,j) / sqrt(sum(W(:,j).^2)); 
         
        change = change + sum(abs(W(:,j) - old_W)); 
         
        if (plot_on > 0), 
            %Assign each of the patterns to a center 
            dist        = W'*x; 
            [m, label]  = max(dist); 
            centers     = zeros(c,N); 
            for i = 1:N, 
                in = find(label == i); 
                if ~isempty(in) 
                    centers(:,i) = mean(x(1:2,find(label==i))')'; 
                else 
                    centers(:,i) = nan; 
                end 
            end 
             
            %Plot the centers during the process  
            plot_process(centers, plot_on) 
        end 
 
    end 
 
    eta = eta * decay; 
     
    if (change/r < 1e-4), 
        break 
    end 
     
end 
 
if (i == max_iter), 
   disp(['Maximum iteration (' num2str(max_iter) ') reached']); 
else 
    disp(['Finished after ' num2str(i) ' iterations.']) 
end 
 
%Assign each of the patterns to a center 
dist        = W'*x; 
[m, label]  = max(dist); 
patterns     = zeros(c,N); 
for i = 1:N, 
    in = find(label == i); 
    if ~isempty(in) 
        patterns(:,i) = mean(x(1:end-1,find(label==i))')'; 
    else 
        patterns(:,i) = nan; 
    end 
end 
 
%Label the points 
[m,label] = min(dist); 
targets   = zeros(1,N); 
Uc        = unique(train_targets); 
for i = 1:N, 
    n           = hist(train_targets(:,find(label == i)), Uc); 
    [m, max_l]  = max(n); 
    targets(i)  = Uc(max_l); 
end