机器学习中利用牛顿迭代法代替梯度下降

function  [theta, J_history] =newtownmethod(X, y, theta, num_iters)
    m = length(y); % number of training examples
    n = size(X,2); % number of training examples
    H=zeros(n,n);
    J_history = zeros(num_iters, 1);
    for iter = 1:num_iters
        h = X*theta;
        for i=1:n
            for j=1:n
                    H(i,j)=sum((X(:,i).*X(:,j)));    
            end
        end
        theta = theta - pinv(H)*X'*(h-y);
    end
end

主要的难点在求黑赛矩阵，在样本不大时，利用牛顿迭代法收敛速度比梯度下降法快得多