Is it possible to make create custom InitialPopulation for N simulation iterates of genetic algorithm w/o resetting optimoptions N times?

I would like to use the genetic algorithm to simulate N nonlinear minimization problems with nonlinear constraints. As the problem is constrained, I get better results when I am able to feed the genetic algorithm a custom creation function/initial population. Is there a way to create N custom subpopulations without resetting optimoptions N times?

Further background/context: I had hoped to use the custom creation function to index the evaluation of the creation function to the particular simulation iteration of the objective function; i.e., on simulation j = 542 of 2,000,000, call a creation functio that accepts the index j as a variable so that the created index j facilitates the creation of a new initial population, something like the following:

function Population = myfun(K,obj2,options,START,ilb,iub,j,G)

Population = (repmat(ilb(j,:),G,1)+repmat(iub(j,:)-ilb(j,:),G,1).*START);

where iub, ilb are custom bounds for each simulation iterate j, G the populzation size, K the dimension of a population member, and START a G-by-K matrix of preallocated uniform random variables.

Thus far, I haven't had any luck. Using the the for-loop index j is no good, as creating a function handle for the creation function freezes the for-loop index at whatever variable existed at the time the handle was created. Additionally, one cannot seemingly make the for-loop index a variable of the creation function, as the genetic algorithm seems to only know how to call on creation functions with the variables (GenomeLength, FitnessFcn, options). I would appreciate any insight available on this matter.

Answer 1

[Alan Weiss]

You can use a custom creation function. As you probably know, this function can have the syntax

function Population = myfun(GenomeLength, FitnessFcn, options,params)

where params can be a parameter vector or structure. You set the function in ga using

options = optimoptions("ga","CreationFcn",...
    @(GenomeLength, FitnessFcn, options)myfun(GenomeLength, FitnessFcn, options,params));

You can set params as you like. But, as you say, you might need to recreate these options each time you change params.

You can also check out using a nested function for passing parameters in Passing Extra Parameters. That way you would not have to call optimoptions each time you update the parameters.

Alan Weiss

MATLAB mathematical toolbox documentation

[Nick Rowe]

Thank you!