MATLAB: How is it possible to pass simulannealbnd an objective function that incorporates data

maximum likelihoodoptimizationsimulannealbndsimulated annealing

I am trying to use simulated annealing (Global Optimization Toolbox) for the MLE estimation of a time-varying copula density. Traditional optimization solvers (e.g., fminsearch and fminunc) easily allow for the incorporation of data in the process. I cannot seem to find a straightforward way to do that using the simulannealbnd function.

For example, I have something like

X0=Initial_Vals;
ObjectiveFunction=@copulalikelihood;
[parameters,Loglikelihood,~,~]=simulannealbnd(ObjectiveFunction,X0);

Any suggestions or insights would be greatly appreciated.

Best Answer

parameterize the function

Related Solutions

MATLAB: Matlab optimization module problem

If you have access to the Global Optimization Toolbox, you may use “GlobalSearch” or “MultiStart” features of the toolbox to obtain the global minimum to your problem.

https://www.mathworks.com/help/releases/R2016b/gads/globalsearch-class.html

https://www.mathworks.com/help/releases/R2016b/gads/multistart-class.html

In addition, another potential workaround is to use the function “patternsearch”

Example:

>> options = optimoptions('patternsearch','Display','iter','MaxFunctionEvaluations',1500,'MaxIterations',1500,'ConstraintTolerance',1.0e-10,'StepTolerance',1.0e-10);

>> estparam=patternsearch(@objfun,estparam0,A,b,Aeq,beq,lb,ub,[],options)

https://www.mathworks.com/help/releases/R2016b/gads/patternsearch.html

Finally, you may also try calling “fmincon” multiple times in a “for” or “parfor” loop, giving different initial guesses each time and keeping track of the best solution. This will not necessarily guarantee a global minimum, but it will likely yield a better and more consistent answer.

MATLAB: Writing Constaints for simulannealbnd

The constraint you have is not a bound, it is a linear inequality constraint that cannot be represented by a bound.

I strongly suggest that you do not use simulannealbnd for this problem, and instead use fmincon if your objective function is smooth, and use patternsearch otherwise. If you are looking for a global solution, then start the solver from a variety of initial points

x0 = lb + rand(size(lb)).*(ub - lb);

By the way, you wrote that your lower bound is x >= 1, not x >= 0, since you said lb = ones and not zeros.

Alan Weiss

MATLAB mathematical toolbox documentation

Best Answer

Related Solutions

MATLAB: Matlab optimization module problem

MATLAB: Writing Constaints for simulannealbnd

Related Question