svc_fun.m

核方法＆svm是模式识别是很重要的方法

function outPutPar=SVC_TwoClass(class_one,class_two,sizeTrainClassOne,sizeTrainClassTwo,kPar,C,kType)
% the function for generating the two_class based classification
% InputPar:
% class_one: the first class data;
% class_two: the second class data;
% sizeTrainClassOne: the size of the training number of the first class
% data;
% sizeTrainClassTwo: the size of the training number of the second class
% data;
% OutputPar:
% outPutPar.w: 
% outPutPar.b;
% outPutPar.alpha;
% Written by WangZhe on 2005-04-05;

Y=[ones(sizeTrainClassOne,1);-1*ones(sizeTrainClassTwo,1)];
% input data completed

X=[class_one(1:sizeTrainClassOne,:);class_two(1:sizeTrainClassTwo,:)];

mat_kernel=Kernel(X,X,kType,kPar);
% calculate the kernel use a kernel

trainSampleNum=size(X,1);
alpha=zeros(trainSampleNum,1);
b=0;
% set the value of alpha ,b and C

littleValue=0;
% the little value control the precison of this program
againstKKTAllowed=0;
% the allowed times that is against KKT conditions
loop_times=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

while(loop_times<5000)
    
    % Step1:Choosing two alphas of samples
    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    index=randperm(trainSampleNum);
    i=index(1); j=index(2);
    % Here using the technique of randomly selecting i,j
    
    if(Y(i)~=Y(j))
        U=max(0,alpha(j)-alpha(i));
        V=min(C,C-alpha(i)+alpha(j));
    else
        U=max(0,alpha(i)+alpha(j)-C);
        V=min(C,alpha(i)+alpha(j));
    end
    % Compute the value of U and V
    
    E1=(alpha.*Y)'*mat_kernel(:,i)+b-Y(i);
    E2=(alpha.*Y)'*mat_kernel(:,j)+b-Y(j);
    
    k=mat_kernel(i,i)+mat_kernel(j,j)-2*mat_kernel(i,j);
    if abs(k)<littleValue+10^(-3)
        continue;
    end;
    
    old_alpha2=alpha(j);
    new_alpha2=old_alpha2+Y(j)*(E1-E2)/k;
    
    if(new_alpha2>V)
        new_alpha2=V;
    else
        if(new_alpha2<U)
            new_alpha2=U;
        end
    end
    
    alpha(j)=new_alpha2;
    alpha(i)=alpha(i)+Y(i)*Y(j)*(old_alpha2-new_alpha2);
    % the modification of alphas have been completed
    
    
    [index value]=find(alpha>littleValue & alpha<C-littleValue);
    if(length(index)>=1)
        b_vector=Y(index)-((alpha.*Y)'*mat_kernel(:,index))';
        b=sum(b_vector)/length(index);
    end
    % reset the value of b
    
    % Step2: check KKT condition
    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    temp=Y(index)'.*(b+(alpha.*Y)'*mat_kernel(:,index));
    against_KKT2=sum(abs(temp-1)>littleValue);
    
    [index value]=find(abs(alpha-C)<littleValue);
    temp=Y(index)'.*(b+(alpha.*Y)'*mat_kernel(:,index));
    against_KKT3=sum(temp>1+littleValue);
    
    [index value]=find(alpha<=littleValue);
    temp=Y(index)'.*(b+(alpha.*Y)'*mat_kernel(:,index));
    against_KKT1=sum(temp<1-littleValue);
    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    
    loop_times=loop_times+1;
    
    if(against_KKT1+against_KKT2+against_KKT3 <= againstKKTAllowed)
        break;
    end    
end

numberSV=length(find(abs(alpha)>littleValue));
outPutPar.w=((alpha.*Y)'*X)';
outPutPar.b=b;
outPutPar.alpha=alpha;