lms_classifier.m

包含用lms、mse、perceptron准则函数的二类分类器

function a = lms_classifier(feature1, feature2, theta, eta)

%-----------------------------------------------------
% a = lms_classifier(feature1, feature2, theta, eta)
% Function calculates the weight vector for the linear discriminant function
% for the two-category data. Algorithm is based on least-mean squared rule.
% Input variables:
% - feature1 -  augmented feature vector for the first class
% - feature2 -  augmented feature vector for the second class
% - theta - threshold (stopping criterion)
% - eta - learning rate 
% Output:
% - a - Weight vector (a_0,a_1,a_2,...,a_d)' for the linear discriminant of the form: 
%   g(x) = a_0 + a_1*x1 + a_2*x2 + ... + a_d*xd 


% --------------------------------------------------
% Evgeny Krestyannikov
% [e-mail]
% Institute of Signal Processing
% Room TE 313



% normalized augmented feature vector
feature2=-feature2;
Y = [feature1 feature2];
sizeY=size(Y,2);
order=randperm(sizeY);
Y=Y(:,order);
b1=ones(1,sizeY);
b2=ones(1,sizeY);
b=[b1 b2]; % vector of margins


% Initialization
a=[1 rand(1,2)]';
k = 0;
eta1 = eta;
update = 1e3;


% LMS algorithm
while (sum(abs(update)) > theta && (k<(sizeY-1)))     
    k = mod(k+1,sizeY); % k=(k+1)mod sizeY 余数
    eta(k)= eta1/k;
    update	= eta(k)*(b(k) - a'*Y(:,k))*Y(:,k) ;
    a	    = a + update;
    
end

    disp(['Did ' num2str(k) ' iterations']);