%訓練過程 Logistic_solve.m

function [beta,rate] = Logistic_solve(x,y,step_in)  %回歸系數beta，正確率rate
step = step_in;
x_training = x;
y_training = y;
s = size(x_training);   %s(1)=N 樣本數
p = zeros(s(1),1);  %初始化p
beta = zeros(s(2),step+1);  %初始化beta=0
W = zeros(s(1));    %初始化W=1
L=0;
for t=1:step    %第2步后,第3步無解，故1步后的beta就是最終解
    for i=1:s(1)
        p(i,1) = exp(beta(:,t)'*x_training(i,1:s(2))') / (1+exp(beta(:,t)'*x_training(i,1:s(2))'));
        W(i,i) = p(i,1)*(1-p(i,1));
    end
    z = x_training*beta(:,t) + inv(W)*(y_training - p);
    beta(:,t+1) = inv(x_training'*W*x_training) * x_training'*z;
end
beta=beta(:,t); %？？？

%標準化y,算誤差點數
for i=1:s(1)
    if p(i)>0.5
        p(i)=1;
    end    
end
L = sum(abs(p - y_training));  %若判斷正確，則p == y_training，L==0 
rate = (s(1)-L)/s(1)*100;   %正確率