Problem with fminunc, why does the parameters not change?

Hi,

may be is a very simple mistake but I cant see it. I have a simple model to try how fminunc works and the thing is I dont get different parameters from the initial ones.

this is my code:

function main2()

pars0=[0.2 0.3 0.4]';

objective=@(x)li2(x);
[x,fval,exitflag,output,grad,hessian]= fminunc(objective,pars0);

disp(x);
disp(fval);
disp(exitflag);

end

function [ilik]=li2(x);

mData=randn(50,1);
iT=size(mData,1);

lik=zeros(iT,1);
hxt=zeros(iT,1);

Yt=0;
xt=0;

for i = 1 : iT
Yt=mData(i,1);
F=x(3,1)*Yt^2+x(1,1)*Yt+x(2,1);
hxt(i,1)=F;
lik(i,1) = -(0.5)*iT*log(2*pi)-0.5*F;
end

ilik = -sum(lik);

end

Thanks a lot in advance.

Answer 1

[Matt J]

That is not how your code behaves for me. I do see a small change between pars0 and the final solution:

>> x-pars0

ans =

     1.0e-05 *

      0.0872
     -0.0150
      0.1102

However, it makes no sense to have randn() statements in your objective function. Doing so means the objective function is not a deterministic function of the input x. How can the iterative minimization algorithm know what to minimize if the definition of the function is changing randomly with every iteration?

[Matt J]

You probably meant to do something like the following. Notice that it keeps the random data fixed while the minimization is in progress. However, without constraints, the minimum is simply -Inf.

pars0=[0.2 0.3 0.4]';

mData=randn(50,1);
objective=@(x)li2(x,mData);
[x,fval,exitflag,output,grad,hessian]= fminunc(objective,pars0);
disp(x);
disp(fval);
disp(exitflag);

keyboard

function [ilik]=li2(x,mData)

iT=size(mData,1);
lik=zeros(iT,1);
hxt=zeros(iT,1);
Yt=0;
xt=0;
for i = 1 : iT
Yt=mData(i,1);
F=x(3,1)*Yt^2+x(1,1)*Yt+x(2,1);
hxt(i,1)=F;
lik(i,1) = -(0.5)*iT*log(2*pi)-0.5*F;
end
ilik = -sum(lik);

Answer 2

[Barbara]

Dear Matt J

Yes the mistake was that I put the random generator of the data inside the function. I see it now. Thanks!