function  [ROCarea,alfa,b,e,Yhat,se]=lssvr(Xs,Ys,gam,sig2)

 % LS-SVM regression + ROC curve
 % Xs: N x #inputs
 % Ys: N x 1
 % N = number of training data
 % gam = regularization constant
 % sig2 = sig^2 for RBF kernel
 
   Xst=Xs';
	N=length(Ys);
   K=zeros(N);
    for i=1:N
       for j=1:i
%        u=Xst(:,i); v=Xst(:,j); u=u';v=v';
%        K(i,j) = exp(-sqrt((u-v)*(u-v)')/(2*sig2))     
        K(i,j)=exp(-(norm(Xst(:,i)-Xst(:,j))^2)/sig2);
        K(j,i)=K(i,j);
      end
   end

   Ic=eye(N);vec1=ones(N,1);

   M=[0 vec1';vec1 (K+Ic/gam)]; 
   alfab=M\[0;Ys];

   alfa=alfab(2:length(alfab));
   b=alfab(1);

   e=alfa/gam;
   Yhat=Ys-e;
 
   Yg=[];
   Yk=[];
   for i=1:length(Ys)
     if sign(Ys(i)) <= 0      
       Yg=[Yg;Yhat(i)];
     else
       Yk=[Yk;Yhat(i)];
     end
   end
   
   [TN,TP,FP,FN,opp,se]=roc_tf(Yg,Yk);
   
   ROCarea=opp;