% N - numero de amostras
% ne - numero de atributos 
function w=rlogb(X,Y)
[N,ne]=size(X);
X=[ones(N,1),X];
w=rand(ne+1,1)/5;
E=1/N*sum(log(1+exp(-(X*w).*Y)));
Emin = 1e-3;
nit = 20;
it =0;
lamb= 0.7;   %taxa de aprendizado
while E>Emin & it<nit
    
    % Incrementa iteração
    it = it+1;  
    
    % Calcula o gradiente
    dEw=zeros(ne+1,1);  
    for j=1:ne+1
        for n=1:N
            dEw(j,1)= dEw(j,1)+Y(n,1)*X(n,j)/(1 + exp(Y(n,1)*X(n,:)*w));
        end
    end
    dEw = -dEw/N;
  
    % atualiza w
    w = w - lamb*dEw;
    
    % calcula o erro
    E=1/N*sum(log(1+exp(-(X*w).*Y)))
    
    pause(0.2)
end