gamultiobj output function error

Hi,

I have a multiobjective opimization problem and want to solve it with gamultiobj. I want to save the optimized values of my objective functions after every iteration in an array. I found the example code for the output function here: http://de.mathworks.com/help/optim/ug/output-functions.html . But when I apply the example code to my problem, like this:

 function [history,searchdir] = runfmincon
 FitnessFunction = @needfunction;
 numberOfVariables = 8;
 lb = zeros(1,8);
 ub = [];
% Set up shared variables with OUTFUN
 history.x = [];
 history.fval = [];
 searchdir = [];

 options = gaoptimset('PlotFcns',@gaplotpareto,'UseParallel',true,'OutputFcns', @outfun,'PopulationSize',5);

 xsol = gamultiobj(FitnessFunction,numberOfVariables,[],[],[],[],lb,[],options);
 function stop = outfun(x,optimValues,state)
     stop = false;

     switch state
         case 'init'
             hold on
         case 'iter'
         % Concatenate current point and objective function
         % value with history. x must be a row vector.
           history.fval = [history.fval; optimValues.fval];
           history.x = [history.x; x];
         % Concatenate current search direction with 
         % searchdir.
           searchdir = [searchdir;... 
                        optimValues.searchdirection'];
           plot(x(1),x(2),'o');
         % Label points with iteration number and add title.
         % Add .15 to x(1) to separate label from plotted 'o'
           text(x(1)+.15,x(2),... 
                num2str(optimValues.iteration));
           title('Sequence of Points Computed by fmincon');
         case 'done'
             hold off
         otherwise
     end
 end

 end

I get the following error:

 Error using runfmincon/outfun
 Too many output arguments.

 Error in gaoutput (line 39)
    [state,optnew,changed] = feval(functions{i},options.OutputPlotFcnOptions,state,flag,args{i}{:});

 Error in gamultiobjsolve (line 13)
 [state,options] = gaoutput(FitnessFcn,options,state,currentState);

 Error in gamultiobj (line 274)
 [x,fval,exitFlag,output,population,scores] = gamultiobjsolve(FitnessFcn,nvars, ...

 Error in runfmincon (line 16)
 xsol = gamultiobj(FitnessFunction,numberOfVariables,[],[],[],[],lb,[],options);

It would be great, if someone could help me out

Answer 1

[Alan Weiss]

The genetic algorithm has a different syntax for output functions than Optimization Toolbox functions. The syntax is described here. After you correct your syntax things will most likely work (I didn't check in detail, but everything else looks reasonable).

Alan Weiss

MATLAB mathematical toolbox documentation

[Maximilian Ernst]

Hi I have now changed the sytnax to:

 function state = outfun(x,optimValues,state)

 optchanged = false;

 switch flag
 case 'init'
        disp('Starting the algorithm');
    case {'iter','interrupt'}
        disp('Iterating ...')
    case 'done'
        disp('Performing final task');
         % Concatenate current point and objective function
         % value with history. x must be a row vector.
           history.fval = [history.fval; optimValues.fval];
           history.x = [history.x; x];
         % Concatenate current search direction with 
         % searchdir.

           searchdir = [searchdir;... 
                        optimValues.searchdirection'];
           plot(x(1),x(2),'o');
         % Label points with iteration number and add title.
         % Add .15 to x(1) to separate label from plotted 'o'
           text(x(1)+.15,x(2),... 
                num2str(optimValues.iteration));
           title('Sequence of Points Computed by fmincon');
         case 'done'
             hold off
         otherwise
     end
 end

But the same error still occurs. I just don't know what I am doing wrong.

[Walter Roberson]

The syntax is

[state,options,optchanged] = myfun(options,state,flag)

but you only have state on the left hand side, so you are only defining one output argument in a context that needs three output arguments.

[Maximilian Ernst]

Thats true and if I change the syntax to

 [state,options,optchanged] = myfun(options,state,flag)

The error does not occur anymore. But then I get an error

Undefined variable "optimValues" or class "optimValues.fval".

what makes sense because I don't pass optimValues anymore to the output function. How can I avoid that error? I just want to store the optimal values after each iteration in the array 'history'.

[Alan Weiss]

The error probably comes from this line of code:

history.fval = [history.fval; optimValues.fval];

You will probably also get an error from the line

history.x = [history.x; x];

Instead, use the appropriate values that gamultiobj passes, namely, state.Population. If you want the fitness function of the population, I think that you need to calculate it yourself. You can get the state.Score matrix, but it is up to you to interpret the scores.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[Maximilian Ernst]

Thank you for your answer. I was now able to save the scorematrix. I created a file that was update in every iteration by the values of state.Score. I also plotted the pareto front, but the last row of my score matrix does not have the same entries as shown in the pareto front, do you know why that could be?

[Alan Weiss]

1. Scores are not the same as objective function values. They are scaled fitness. I told you that you would have to interpret the scores.
2. The Pareto front is only some of the members of the population, not the entire population.

Alan Weiss

MATLAB mathematical toolbox documentation

[Maximilian Ernst]

Thank you very much for your answer. May I ask, if there is a way to get the objective function values from the score? I have read the documentation about scaled fitness and there it says the objective function values are somehow proportional to the score by a factor sqrt(r),where r is the rank. Can this be used to obtain the objective function values from the scores? Or is there anaother way to get the objective function values?

[Alan Weiss]

You can get the entire population at each iteration from the state.Population field. You can then get the fitness function values by evaluating the fitness functions on the population. There is no other way, as I already told you: "If you want the fitness function of the population, I think that you need to calculate it yourself."

Alan Weiss

MATLAB mathematical toolbox documentation

[Maximilian Ernst]

ok thank you very much

Answer 2

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/301400-extra-paramters-to-ga-output-function

[Maximilian Ernst]

Hi, I read your answer but I still don't know, what I did wrong. I am using a nested output function as you suggested. Actually I am using almost exactly the same code as in the mathworks example which link I posted in my question. The example works fine, so maybe it has something to do with the different solver? (I am using gamultiobj and not fmincon like in the example)

Answer 3

[Modess Seyednezhad]

Hi,

I have the same issue. The syntaxt page was not useful for me. I could not find my error.

function [history,searchdir] = runmultiobjfunc

% Set up shared variables with OUTFUN

history.x = [];

history.fval = [];

searchdir = [];

% call optimization

nvar = 2;

Lb = [1, 3];

Ub = [19.2, 6.27];

options = optimoptions(@gamultiobj,'OutputFcn',@outfun,...

'Display','iter', 'MaxGenerations',200,'PopulationSize',50);

xsol = gamultiobj(@objfun,nvar,[],[],[],[],Lb,Ub,options);

% Set up output function

function stop = outfun(x,optimValues,state)

stop = false;

switch state

case 'init'

hold on

case 'iter'

% Concatenate current point and objective function

% value with history. x must be a row vector.

history.fval = [history.fval; optimValues.fval];

history.x = [history.x; x];

% Concatenate current search direction with searchdir.

searchdir = [searchdir;...

optimValues.searchdirection'];

plot(x(1),x(2),'o');

% Label points with iteration number and add title.

% Add .15 to x(1) to separate label from plotted 'o'

text(x(1)+.15,x(2),...

num2str(optimValues.iteration));

title('Sequence of Points Computed by gamultiobj');

case 'done'

hold off

otherwise

end

end

end

The error is:

Error using runmultiobjfunc/outfun

Too many output arguments.

Error in gaoutput (line 39)

[state,optnew,changed] = feval(functions{i},options.OutputPlotFcnOptions,state,flag,args{i}{:});

Error in gamultiobjsolve (line 13)

[state,options] = gaoutput(FitnessFcn,options,state,currentState);

Error in gamultiobj (line 303)

[x,fval,exitFlag,output,population,scores] = gamultiobjsolve(FitnessFcn,nvars, ...

Error in runmultiobjfunc (line 15)

xsol = gamultiobj(@objfun,nvar,[],[],[],[],Lb,Ub,options);