MATLAB: Vary step size for fminunc

fminuncgradientpremature convergencestep size

Is there a way to adaptively update DiffMinChange during execution of fminunc? I ask because I am using fminunc in a problem where it's sometimes converging to the initial point, which is not the minimum, because it does not use a large enough step-size in evaluating the gradient so the function does not appear to change.

I thought it may be possible using an output function, but from what is says here : "Using an output function you can view, but not set, optimization quantities", it appears this may not be the case. Has anyone tried this before, or have any suggestions?

Thanks,

Best Answer

Sorry, no, you cannot set DiffMinChange adaptively. You might want to use the FinDiffRelStep option instead, but I don't know that this will really help.

As the note in the link states, it is hard to know how to set these finite differencing options.

Alan Weiss

MATLAB mathematical toolbox documentation

Related Solutions

MATLAB: Matlab: Get the Hessian matrix using Multistart

MultiStart does not return the Hessian. But, if you really want the Hessian for computing confidence intervals, you shouldn't use the fmincon Hessian anyway, because, as documented, it is inaccurate.

I do not know what the confidence interval calculation is when there are active constraints in a maximum likelihood calculation. If, at the end of the calculation, there are no active constraints, then you can call fminunc on the function at the final point. As documented, fminunc Hessians are accurate.

Alan Weiss

MATLAB mathematical toolbox documentation

MATLAB: Using fminunc()

what do you want? what is the function you want to optimize for? fminunc searches for a minimum, but for instance x1+x2 is unbounded below (solution: x1=-inf, x2=-inf) and for sin(x3) has infinitely many minima...

anyway the format for fminunc (and fmincon) is the following:

myfunction=@(x)x(1)+x(2)+sin(x3);
x0=[1, 4, 10];
[x,y]=fminunc(myfunction,x0);

note that the my example does not have a solution...

Best Answer

Related Solutions

MATLAB: Matlab: Get the Hessian matrix using Multistart

MATLAB: Using fminunc()

Related Question