returning additional values from a fitness function for genetic optimisation

Is there a simple way to return additional values from a fitness function (in addition to the objective value) when using an optimisation algorithm, such as genetic algorithm or multi-objective ga in matlab?

For example, it is easy enough to specify additional parameters to be passed to a fitness function using a function handle:

f = @(x)parameterfun(x,a,b,c);

However, is possible to do something similar for the outputs of the fitness function?

Some ideas I had include saving the values in the fitness-function to a file, and then reading them in after the ga has finished in the parent function, or saving the values as global variables, but is there a simpler approach?

Answer 1

[Alan Weiss]

You can use nested functions to do what you want. The example of a nested output function might help.

Alan Weiss

MATLAB mathematical toolbox documentation

[A Desai]

Hi Alan, I've taken a look at the nested output function example but I don’t understand how it would work. I have the ga working currently but want to log all the outputs from the fitness function at the end of each generation for more analysis.

I have the following parameters passed to the fitness function, and the line for the GA :

f=@(x)framework0(x,y,z);

[DVbest, sold, exit, output, population, scores] = ga(f,14,[],[],[],[],DV_l,DV_u,[],options);

The fitness function, framework0, has 3 outputs I want to log, the function looks something like this:

function [output] = framework0(x,y,z)

[x1,y1,z1] = regression (x,y,z);

[x2,y2,z2] = framework1 (x,y,z,x1,y1,z1);

output = -x2(3); %Single value output to GA

end

I had a similar idea to save the output values (x2,y2,z2) in the fitness function but I’m assuming using nested functions would be quicker to operate.

[Alan Weiss]

Usually, people use an output function to record what they like.

Alan Weiss

MATLAB mathematical toolbox documentation

[A Desai]

I currently have an output function which records the population, score and state - but i dont understand how to pass through the extra variables from the fitness function to the output function.

[Alan Weiss]

You pass extra variables to an output function in exactly the same way that you pass extra parameters anywhere else.

Alan Weiss

MATLAB mathematical toolbox documentation

[A Desai]

Do you mind showing me what that would look like using the above sample code? I understand passing the extra parameters in the example provided in the link, which is similar to passing variables into the fitness function, but i cant seem to get it to work from the ga to the output function. From what I understood so far, I've added the following to the options line:

options = optimoptions('ga','CreationFcn',@gacreationframework,...
'OutputFcn',@(options,state,flag)gaoutputframework(options,state,flag,fout),...
'MaxGenerations',1,...
'PopulationSize',4,...
'UseVectorized',false,...
'UseParallel',true); %

and followed a similar line with the output function:

function [state,options,optchanged] = gaoutputframework(options,state,flag,fout)

where fout contain the extra parameters within the fitness function. But this doesnt work...