arrays - 在 Matlab 中使用匿名 fnc 时如何将值存储到数组中以绘制它们

Question

我在绘制迭代次数与成本函数（f_star1）时遇到问题。我无法将迭代（niter）和在每次迭代中评估的函数保存到数组中。下面是我的代码。while 循环中的注释中指定的行是我卡住的地方！

% define initial guess
x = [1 1]'; 

% define  parameters
h = 1;          % initial step size 
beta = 0.7;     % Armijo's Rule: Beta  = (0,1)
alpha = 0.2;    % Armijo's Rule: Alpha = (0,.5)
grad = 1;       % initialize gradient 
epsilon = 1e-2;
niter = 0;      % iterations counter
f = @costf;     % objective fcn assigned to variable f
i = 1;
n=0;
% Loop
while norm(grad) > epsilon
    grad = gradient(x(:,end));

    while f(x(:,end)-h*grad) > (f(x(:,end))-alpha*h*grad) %backtracking line search 
        h = h*beta;
    end
    x(:,end+1) = x(:,end) - h*grad;

    %This is the part I am stuck!!!
    x1G = x(1,:)';
    x2G = x(2,:)';
    x_star1 = [(x(1,:));(x(2,:))]';
    f_star1(i,:) = costf(x_star1(i,:));

%     figure(1), plot(iterations,f_star1,'*')

    niter= niter+1;

end
%      plot(f_star1,niter,'*')


%Cost Function
function f = costf(x)
    f = exp(x(1)+3*x(2)-0.1) + exp(x(1)-3*x(2)-0.1) + exp(-x(1)-0.1);
end

% find gradient with central difference
function df = gradient(x0)
    f = @costf; % assign cost fcn function to variable f
    h_step = .01; % perturbation step
    dx1 = (f(x0+[h_step;0])-f(x0-[h_step;0]))/(2*h_step); %slope
    dx2 = (f(x0+[0;h_step])-f(x0-[0;h_step]))/(2*h_step);
    df = [dx1;dx2]; 
end

标签： arraysmatlabplotanonymous-function

Answer 1

首先，要使用plot(niter, f_star1)函数，niter并且f_star1在您的情况下应该是两个长度相等的向量。问题i出在while循环中。你需要i在迭代中增加。

% define initial guess
x = [1 1]'; 

% define  parameters
h = 1;          % initial step size 
beta = 0.7;     % Armijo's Rule: Beta  = (0,1)
alpha = 0.2;    % Armijo's Rule: Alpha = (0,.5)
grad = 1;       % initialize gradient 
epsilon = 1e-2;
niter = 0;      % iterations counter
f = @costf;     % objective fcn assigned to variable f
i = 1;
n=0;
% Loop
while norm(grad) > epsilon
    grad = gradient(x(:,end));

    while f(x(:,end)-h*grad) > (f(x(:,end))-alpha*h*grad) %backtracking line search 
        h = h*beta;
    end
    x(:,end+1) = x(:,end) - h*grad;

    %This is the part I am stuck!!!
    x1G = x(1,:)';
    x2G = x(2,:)';
    x_star1 = [(x(1,:));(x(2,:))]';
    f_star1(i,:) = costf(x_star1(i,:));

    niter(i,:)= i-1;
    i = i+ 1;

end
plot(niter,f_star1,'*-')
title('f\_star1 vs niter');
xlabel('niter');
ylabel('f\_star1');


%Cost Function
function f = costf(x)
    f = exp(x(1)+3*x(2)-0.1) + exp(x(1)-3*x(2)-0.1) + exp(-x(1)-0.1);
end

% find gradient with central difference
function df = gradient(x0)
    f = @costf; % assign cost fcn function to variable f
    h_step = .01; % perturbation step
    dx1 = (f(x0+[h_step;0])-f(x0-[h_step;0]))/(2*h_step); %slope
    dx2 = (f(x0+[0;h_step])-f(x0-[0;h_step]))/(2*h_step);
    df = [dx1;dx2]; 
end

输出：