Passing extra outputs from cost function

Dear Matlab Community,

I am using a GA to optimise a certain function, but I want to keep saving some additional data.

f=@(x)requestFitting(extractionInit,x); problemMultiga.fitnessfcn=f;

with

[f,extractionFit] = requestFitting(extractionInit,x)

The problem is, that through the function handle, information on the output of the cost function:extractionFit is lost.

Please help me, how can I save additional outputs of the cost function?

Thanks a lot!

 [cost,]

Answer 1

[Alan Weiss]

There are several ways. Proper use of nested functions is one way. Using an output function is another.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[Stefano Coss]

Thanks Alan, well I have tried to use those, but somehwo I dont get it right. How must the design of the nested functin look like.? Could you give me a more detailed hint how it should look like?

Thanks!

[Alan Weiss]

Take a look at this example, which uses a nested function to extract information and store it. Or take a look at this example of a nested output function.

Yes, these are complicated, subtle examples. For your case you will have to look at the overall idea and somehow adapt it.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[Kurt Stewart]

Hi Alan,

I am wondering what the advantage of using nested functions approach over global variables? I understand functions would have their own local stacks is the only reason to keep it cleaner? Essentially they are 'global' variables for the function and nested functions. I am wondering if I could run an optimizer in parallel with this method. Thanks. I used the example you referenced and adapted for gamultiobj.

[Kurt Stewart]

This seemed to work - a mashup of examples

function [x Fval rand_value_mat] = myproblem3(x0)

rand_value=[];
rand_value_mat=[];
states=[];
final_state=[];
FitnessFunction = @kurt_multiobjective; % Function handle to the fitness function
numberOfVariables = 3; % Number of decision variables
lb = [-5 -5 -5]; % Lower bound
ub = [5 5 5]; % Upper bound
A = []; % No linear inequality constraints
b = []; % No linear inequality constraints
Aeq = []; % No linear equality constraints
beq = []; % No linear equality constraints
options = optimoptions(@gamultiobj,'PlotFcn',@gaplotpareto,'OutputFcn', @myoutput,'UseParallel',true);

%%
% Run the |gamultiobj| solver and display the number of solutions found on
% the Pareto front and the number of generations.
[x,Fval,exitFlag,Output] = gamultiobj(FitnessFunction,numberOfVariables,A, ...
    b,Aeq,beq,lb,ub,options);

for i=1:18
    index(i)=find(final_state.Score==Fval(i,:),1)
end
Fval(:,3)=final_state.speed(index);

fprintf('The number of points on the Pareto front was: %d\n', size(x,1));
fprintf('The number of generations was : %d\n', Output.generations);

    function[state,options,optchanged] = myoutput(options,state,flag)
        optchanged = false;
        switch flag
            case 'init'
                
            case {'iter','interrupt'}
                rand_value_mat=[rand_value_mat,rand_value];
                states=[states,state];
            case 'done'
                 state.speed=9.83*ones(length(state.Population),1);
                 final_state=[state];
        end
        
    end


    function y = kurt_multiobjective(x)
        % using kur_multiobjective
        % Initialize for two objectives
        y = zeros(2,1);
        rand_value=10;
        % Compute first objective
        for i = 1:2
            y(1) = y(1)  - 10*exp(-0.2*sqrt(x(i)^2 + x(i+1)^2));
        end
        
        % Compute second objective
        for i = 1:3
            y(2) = y(2) +  abs(x(i))^0.8 + 5*sin(x(i)^3);
        end
        
    end
save('workspace.mat')
end

[Kurt Stewart]

I tried parallel and it didnt seem to be a problem, I would have to know more about resource sharing that Matlabs pools do

[Kurt Stewart]

Actually parallel did mess up