How to use Parallel Computing inside ga (genetic algorithm)

Hello everyone,

I´m stuck in my code. I´m trying to maximize a function using genetic algorithm and recently, I read that Parallel Computing could reduce the calculation time that ga takes to show its result, but when I include in gaoptimset('UseParallel',true) the code ends very quickly (at first iteration) and do not optimize my function. I'm using the following code in Matlab:

Note: - Please obviate the first (/if true/ and the last /end/) because I employed {}code to put my code. - the function to optimze is named @func1

I do not know what is wrong in my code because I read in matlab help and all the comments said that if you want to use Parallel Computing you only need to add 'UseParallels',true into gaoptimize.

Thanks for your help.

if true
% The code that I'm using is:
% ***************************
function ga_optim(func)
close all
global fin y_opt
fin = 50000;
y_opt = 9999;
% Number of variable to optimize
nvars = 6;
if nargin == 0
  func = 1;
end
if func == 1
  ObjectiveFunction = @func1;
end
ConstraintFunction = @simple_constraint;
LB = [7; 7; 70; -150; 200; 300];   % Lower bound
UB = [15; 12; 300; 50; 450; 650];  % Upper bound
popsize = 600;
gensize = 600;
opts = gaoptimset('PopulationSize', popsize, 'Generations', gensize, 'Display', 'off', 'TolFun', 1e-2,'UseParallel', true);
t1 = tic;
[x,fval,exitFlag,output,population,scores] = ga(ObjectiveFunction,nvars, [], [], [], [], LB, UB, ConstraintFunction, 1:2, opts);
tiempo = toc(t1);
x = x';
% End of the code
% ***************

end

Answer 1

[Alan Weiss]

What error messages are you seeing, if any? Did you try to create a parallel pool before starting GA? It is possible that you have an older version of Global Optimization Toolbox where you should set UseParallel to 'always'. It is also possible that you do not have a Parallel Computing Toolbox license. Without more details, it is hard to know what might be happening.

Alan Weiss

MATLAB mathematical toolbox documentation

Answer 2

[Camilo]

Hello Alan, I also tried with 'UserParallel','always' but the same thing occurs, GA terminates its executions at first iteration. I do not receive error mensages. There is sb interesting in your comment. How can I create a parallel pool before starting my GA?

[Alan Weiss]

How to use parallel processing has that information. Depending on your software version, and license, this pool can start automatically.

Alan Weiss

MATLAB mathematical toolbox documentation

Answer 3

[Camilo]

I use the following code to start Parallel computing:

opts = gaoptimset('PopulationSize', popsize, 'Generations', gensize, 'Display', 'off', 'TolFun', 1e-2, 'UseParallel', true);

And I've also tried 'UseParallel','always' but the GA ends at first iteration and GA does not find me a suitable optimize set

[Alan Weiss]

Please read the link I gave before about how to start a parallel pool.

Alan Weiss

MATLAB mathematical toolbox documentation

Answer 4

[wonderkismet]

You can use vectorize method.

1. Start a pool
2. In ga options, Enable vectorized
3. process the vectorized generation input with your fitness function. Inside the fitness function, use a parfor to process each row of the generation. The generation is a matrix with population number of rows, segment the rows into the number of works you have and sent them to each work to calculate.

[Terry]

Can you give more details about step 3? It's very complex and hard to understand. Thank you!