How to output additional variables from objective function using ga optimization?

Hi,

I am having trouble to output variables from the objective function using ga other than the value of the objective funtion.

Does anyone have an idea on how to do that?

Greeting,

Moritz

Answer 1

[Alan Weiss]

I am not sure what you are trying to do. You might be trying to pass intermediate values calculated within ga to reuse, for example in a constraint function. I am sorry, but that generally does not work for ga, as the order of operations that ga internally uses does not lend itself to that purpose. In other words, the technique in Objective and Nonlinear Constraints in the Same Function does not work for ga.

If you are trying to do something else, ask again with more detail.

Alan Weiss

MATLAB mathematical toolbox documentation

[Moritz Wegner]

Hi,

thank you for your reply.

I am using ga for the optimization of an electric motor. I am using ga to generate random motor configurations. The objectiv function then calls a finite element programm which returns the mean torque and other parameters which are then used to calculate the value of the fitness function to evaluate the randomly generated motor configuration. I am using an output function which saves the fitness function value of the best individual of each generation and the individual. I am trying to store the parameters such as torque as well. And I didnt find a way to return those parameters from the objective function to pass them to the output function. Is there a way?

Answer 2

[Star Strider]

‘Does anyone have an idea on how to do that?’

After the optimisation terminates, run the objective function with the optimised parameter values and return all the desired outputs.

If the fitness function returns several results, for example:

function [a,b,c] = ftnsfcn(a,x)

... CODE ...

end

and the ga call is something like this:

[A,fval] = ga(@(a)ftnsfcn(a,x), NrParms);

after the optimiization terminates, run this:

[a,b,c] = ftnsfcn(A,x)

This would be easier to illustrate with a more specific example if you have one to share.

.

[Moritz Wegner]

But I want to save all the parameters of every individual in every generation. If I do it the way you propesed (after optimization terminates) it would only return the parameters of the last indivual?

[Star Strider]

It would return the parameters of the best individual, yes.

The only way to save all the values for every iteration of every individual would be to write those to some sort of file every time. The ga optimisation under those circumstnaces could take days, if not longer, to converge, and then you would have the problem of going through all those saved results, only some of which would be meaningful. (I wrote a funciton that would save the results of the best individual in every generation in How to save data from Genetic Algorithm in case MATLAB crashes? - MATLAB Answers - MATLAB Central.) Perhaps getting the parameters from only those saved individuals would be enough. You could do that offline after ga converges.

[Moritz Wegner]

I think i am using a pretty similar approach at the moment. I am saving the data for the best individual od each generation to a file. Then loading the file and adding the data for the next generation. The duration od the optimization isnt the concern, I want to train a neural network with the data later on, so its obly about how to store the data. To my knowledge the approach of saving the data to a file doesnt work with parameters that are only defined in the objective function because the output function cant access those. And that is what I am trying to archieve. I somehow have to pass those parameters to the output function to store them to a file.

[Star Strider]

I’m slightly lost at this point. You can save the best individual in each generation, and then afterwards use that information as input to the fitness function and get all the outputs from it for each saved individual.

That’s what I would do, anyway.

Answer 3

[Walter Roberson]

Have your objective function return a data structure of additional values. memoize() your objective function with a fair size buffer. Have the output function (which runs once per generation) fetch the best value of the model parameters and call the memorized function to retrieve the associated data structure, and store the values somewhere.

[Moritz Wegner]

How does the returning work? Could you maybe give me an example code? I havent found any documentation on the syntax.

[Walter Roberson]

    memoized_objective = memoize(@objective);
    memoized_objective.Cachesize = 1000;
    options = optimoptions('ga','outputfcn', @(options,state,flag)save_best_data(options,state,flag,memoized_objective));
    
    initialize_saved_data();
    
    [bestx, fval, exitflag] = ga(memoized_objective, nvars, A, b, Aeq, beq, lb, ub, nonlcon, options);
    
    best_data = recall_best_data();
    
    function [cost, data_to_save] = objective(x)
       stuff
       data_to_save = {torque, other_parameters};
       cost = whatever;
    end
    
    function initialize_saved_data()
       do whatever is needed to initialize the store of saved data
    end
    
    function [state,options,optchanged] = save_best_data(options,state,flag,memoized_objective)
        if state.LastImprovement == state.Generation
           scores = state.Score;
           [~, minidx] = min(scores);
           best_params = state.Population(minidx,:);
           [best_cost, data_to_save] = memoized_objective(best_params);
           now save data_to_save however is appropriate
        end
    end

    function all_saved_data = recall_best_data()
        now recall the saved data and assign it to all_saved_data
    end