svmmulticlass.m

我用于脑机接口P300拼写范例中的程序

function [xsup,w,b,nbsv,pos,alpha]=svmmulticlass(x,y,nbclass,C,epsilon,kernel,kerneloption,verbose, alphainit)


% USAGE 
%[xsup,w,b,nbsv,pos,alpha]=svmmulticlass(x,y,nbclas,C,epsilon,kernel,kerneloption,verbose, alphainit)
%
% Support vector machine for multiclass CLASSIFICATION
% This routine classify the training set with a support vector machine
% using quadratic programming algorithm (active constraints method)
%
% INPUT
%
% Training set
%      x  		: input data 
%      y  		: output data
% parameters
%		c		: Bound on the lagrangian multipliers     
%		lambda		: Conditioning parameter for QP method
%		kernel		: kernel  type. classical kernel are
%
%		Name			parameters
%		'poly'		polynomial degree
%		'gaussian'	gaussian standard deviation
%
%		for more details see svmkernel
% 
%		kerneloption : parameters of kernel
%
%		for more details see svmkernel
%
% 		verbose : display outputs (default value is 0: no display)
%
%       alphainit : initialization vector of QP problem
%
% OUTPUT
%
% xsup	coordinates of the Support Vector
% w      weight
% b		bias
% pos    position of Support Vector
% alpha  Lagragian multiplier
%
%
% see also svmreg, svmkernel, svmval

%	06/01/2003 Alain Rakotomamonjy
%
%       [e-mail], [e-mail]


if nargin< 7
    alphainit=[];
end;


if nargin < 6
    verbose = 0;
end

if nargin < 5
    kerneloption = 1;
end

if nargin < 4
    kernel = 'gaussian';
end

if nargin < 3
    lambda = 0.000000001;
end

if nargin < 2
    C = 100000;
end

%------------------------------------------------------
% initialisation
%-------------------------------------------------------
[y,ind]=sort(y);
x=x(ind,:);

yextended=repmat(y,nbclass,1);
xextended=repmat(x,nbclass,1);


ell=size(x,1);
n=sum(y==1);

if size(C,1)==1
    C=C*ones(size(y));
end;

%------------------------------------------------------
% construction des matrices associ