% calculate betas solving min. problem (12) in paper

function [beta EXITFLAG] = betaKMM(X, Y, Xtst, Ytst, sigma, doingRealTraining, regression);

nsamples = size(X,1);  % number of train samples
ntestsamples = size(Xtst,1);  % number of test samples

% variables: (here in the program / in (12) in the paper)
% H is K
% f is kappa
% 

% minimize...
'calculating H=K...'    
H = rbf_dot(X,X,sigma,0);
H=(H+H')/2; %make the matrix symmetric (it isn't symmetric before because of bad precision)

'calculating f=kappa...' 
R3 = rbf_dot(X,Xtst,sigma,0);
f=(R3*ones(ntestsamples, 1));
f=-nsamples/ntestsamples*f;

% did the same, but slowlier:
% f=-nsamples/ntestsamples*ones(nsamples,1);
% for i=1:nsamples
%     fi=0;
%     for j=1:ntestsamples
%         fi = fi + rbf_dot(X(i,:),Xtst(j,:),sigma);
%     end
%     f(i,1) = f(i,1)*fi;
% end
% 
% do they really the same? 
%'different f?'
%[f1 f]        

% subject to...
% abs(sum(beta_i) - m) <= m*eps
% which is equivalent to A*beta <= b where A=[1,...1;-1,...,-1] and b=[m*(eps+1);m*(eps-1)]
eps = (sqrt(nsamples)-1)/sqrt(nsamples);
%eps=1000/sqrt(nsamples);
A=ones(1,nsamples);
A(2,:)=-ones(1,nsamples);
b=[nsamples*(eps+1); nsamples*(eps-1)];

Aeq = [];
beq = [];
% 0 <= beta_i <= 1000 for all i
LB = zeros(nsamples,1);
UB = ones(nsamples,1).*1000;

% X=QUADPROG(H,f,A,b,Aeq,beq,LB,UB) attempts to solve the quadratic programming problem:
%              min 0.5*x'*H*x + f'*x   
% subject to:  A*x <= b 
%              Aeq*x = beq
%              LB <= x <= UB 

'solving quadprog for betas...'
[beta,FVAL,EXITFLAG] = quadprog(H,f,A,b,Aeq,beq,LB,UB);
EXITFLAG
if ((EXITFLAG==0 ) && (doingRealTraining==1))
    %[beta,FVAL,EXITFLAG] = quadprog(H,f,A,b,Aeq,beq,LB,UB,beta,optimset('MaxIter',1e4));
    EXITFLAG
end

if (regression==0)
    % guarantee that all beta greater than 0
    threshold=0.01*abs(median(beta));
    beta(find(beta<threshold)) = threshold;
    sprintf('number of beta < %f: %d (0 is good)', threshold, length(find(beta<threshold)))
end


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% EXITFLAG:
%       1  QUADPROG converged with a solution X.
%       3  Change in objective function value smaller than the specified tolerance.
%       4  Local minimizer found.
%       0  Maximum number of iterations exceeded.
%      -2  No feasible point found.
%      -3  Problem is unbounded.
%      -4  Current search direction is not a direction of descent; no further 
%           progress can be made.
%      -7  Magnitude of search direction became too small; no further progress can
%           be made.