Fmincon and loop for updating objective function

Question

Dat Tran 2016-2-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/268475-fmincon-and-loop-for-updating-objective-function

评论： Dat Tran 2016-2-17

采纳的回答： Walter Roberson

在 MATLAB Online 中打开

Dear all,

How can I make a loop for the code below: every step, I want to update objective function in such a way that pr=p.

I would appreciate any thoughts and helps!

Dat

function [p,fval] = MC_NT_try2(p0,Aeq,beq,N)
    if nargin < 5
        opts = optimoptions('fmincon','Algorithm','interior-point','GradObj','on','DerivativeCheck','on');
    end
      M=length(p0);
      p=nan(M,N);
      fval=nan(1,N);
      lb=zeros(64,1);
      ub=ones(64,1);
     for ii=1:N
     [p(:,ii),fval(ii)] = fmincon(@fun,p0,[],[],Aeq,beq,lb,ub,[],opts);
     end
   function [f, gradf]= fun(p)
  %%impose prior information
  if ii==1
  pr1=[0.3185  0.0001  0.1574  0.2902  0.0003  0.00001  0.8426  0.7098  0.6804  0.0822  0.00001  0.00001  0.0008  0.9177  0.00001  0.00001];
  pr2=0.25*ones(64,1);
  else
  pr1=p(:,ii);
  pr2=p(:,ii); 
  end
%%objective function
f1 = 0;
for i = 1:16
    f1 = f1 + p(i)*log(p(i))-p(i)*log(pr1(i));
end
f2=0;
for j = 17:64
    f2 = f2 + p(j)*log(p(j))-p(j)*log(pr2(j));
end
f=f1+f2;
%%gradient of objective function
if nargout > 1
     gradf=[log(p(1))+1-log(pr1(1));
         log(p(2))+1-log(pr1(2));
     log(p(3))+1-log(pr1(3));
     log(p(4))+1-log(pr1(4));
     log(p(5))+1-log(pr1(5));
     log(p(6))+1-log(pr1(6));
     log(p(7))+1-log(pr1(7));
     log(p(8))+1-log(pr1(8));
     log(p(9))+1-log(pr1(9));
     log(p(10))+1-log(pr1(10));
      log(p(11))+1-log(pr1(11));
     log(p(12))+1-log(pr1(12));
     log(p(13))+1-log(pr1(13));
     log(p(14))+1-log(pr1(14));
     log(p(15))+1-log(pr1(15));
     log(p(16))+1-log(pr1(16));
     log(p(17))+1-log(pr2(17));
     log(p(18))+1-log(pr2(18));
     log(p(19))+1-log(pr2(19));
     log(p(20))+1-log(pr2(20));
      log(p(21))+1-log(pr2(21));
     log(p(22))+1-log(pr2(22));
     log(p(23))+1-log(pr2(23));
     log(p(24))+1-log(pr2(24));
     log(p(25))+1-log(pr2(25));
     log(p(26))+1-log(pr2(26));
     log(p(27))+1-log(pr2(27));
     log(p(28))+1-log(pr2(28));
     log(p(29))+1-log(pr2(29));
     log(p(30))+1-log(pr2(30));
      log(p(31))+1-log(pr2(31));
     log(p(32))+1-log(pr2(32));
     log(p(33))+1-log(pr2(33));
     log(p(34))+1-log(pr2(34));
     log(p(35))+1-log(pr2(35));
     log(p(36))+1-log(pr2(36));
     log(p(37))+1-log(pr2(37));
     log(p(38))+1-log(pr2(38));
     log(p(39))+1-log(pr2(39));
     log(p(40))+1-log(pr2(40));
      log(p(41))+1-log(pr2(41));
     log(p(42))+1-log(pr2(42));
     log(p(43))+1-log(pr2(43));
     log(p(44))+1-log(pr2(44));
     log(p(45))+1-log(pr2(45));
     log(p(46))+1-log(pr2(46));
     log(p(47))+1-log(pr2(47));
     log(p(48))+1-log(pr2(48));
     log(p(49))+1-log(pr2(49));
     log(p(50))+1-log(pr2(50));
      log(p(51))+1-log(pr2(51));
     log(p(52))+1-log(pr2(52));
     log(p(53))+1-log(pr2(53));
     log(p(54))+1-log(pr2(54));
     log(p(55))+1-log(pr2(55));
     log(p(56))+1-log(pr2(56));
     log(p(57))+1-log(pr2(57));
     log(p(58))+1-log(pr2(58));
     log(p(59))+1-log(pr2(59));
     log(p(60))+1-log(pr2(60));
      log(p(61))+1-log(pr2(61));
     log(p(62))+1-log(pr2(62));
     log(p(63))+1-log(pr2(63));
     log(p(64))+1-log(pr2(64))];
end

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Walter Roberson 2016-2-16

在 MATLAB Online 中打开

gradf = log(p(:)+1) - log([ reshape(pr1(1:16),[],1); reshape(pr2(17:64),[],1)]);

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2016-2-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/268475-fmincon-and-loop-for-updating-objective-function#answer_210089

在 MATLAB Online 中打开

function [p, fval] = MC_NT_try2(p0, Aeq, beq, N, opts)
  if nargin < 5
        opts = optimoptions('fmincon', 'Algorithm', 'interior-point', 'GradObj', 'on', 'DerivativeCheck', 'on');
  end
    M = length(p0);
    p = nan(M,N);
    fval = nan(1,N);
    lb = zeros(64,1);
    ub = ones(64,1);
    pr = [0.3185, 0.0001, 0.1574, 0.2902, 0.0003, 0.00001, 0.8426, 0.7098, 0.6804, 0.0822, 0.00001, 0.00001, 0.0008, 0.9177, 0.00001, 0.00001, 0.25 * ones(1, 64-16)];
    p0 = p0(:);
    pr = pr(:);
    for ii=1:N
       [pr, fval(ii)] = fmincon(@(p) fun(p, pr), p0, [], [], Aeq, beq, lb, ub, [], opts);
       p(:,ii) = pr;
    end
end
function [f, gradf] = fun(p, pr)
    %%objective function
    f = sum( p .* log(p) - p .* log(pr) );
    %%gradient of objective function
    if nargout > 1
      gradf = log(p+1) - log(pr);
    end
  end