regressão-logistica-binaria

#brew install octave

# impl regressão logistica binaria (sem dimensão y)
# return w-> pesos(x: 'matriz input', y: 'quant desejada')
# n  -> num amostras
# ne -> num entradas / atributos
pt=100
x = [0.5*rand(pt,2);2+0.5*rand(pt,2)]
plot(x(:,1),x(:,2),'*')
y = [ones(pt,1);-ones(pt,1)]
plot(y,'*')

function w=rlogb(x, y)
    [n, ne]= size(x)
    x= [ones(n,1), x]
    w= rand(ne+1, 1)
    e= 1/n * sum(log(1+exp(-(x*w).*y)))
    Emin= 1e-3
    nit= 1000
    it= 0
    lamb= 0.7  #taxa aprendizado
    while e>Emin & it<nit
        #increment interação
        it = it+1
        #calcula gradiente
        dEw=zeros(ne+1,1)
        for j=1:ne+1          #atributo
            for n=1:n         #num padrao / amostra
                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 erro
        e = 1/n * sum(log(1+exp(-(x*w).*y)))
        pause(2)
     end;
end;

rlogb(x,y)