when using genetic algorithm, the number of variables(nvar) is dependant on the row vector(x) that my fitness function accepts. How can I deal with that?

Hello all,

I need to write something like this: ga(h, sum(x(1:5)),[],[],[],[],LB,UB,[],[],ga_opts);

as you see, the number of variables (nvar) is dependant on the vector I want to optimize(x). Is there any way I can deal with this problem?

Thank you

Answer 1

[Evan]

Have you looked into the varargin function?

help varargin

You could then parse the arguments contained within varargin based upon whatever conditions of your row vector x that determine their order/number.

[Shadan]

Thank you for your response. The thing is that varargin will not be recognized outside of the function definition. for example if I write: ga(h, sum(varargin{1}(1:5)),[],[],[],[],LB,UB,[],[1:5],ga_opts);

Answer 2

[amanita]

I dont know if this is relevant, but i usually set nvars as the maximum number of variables and keep in the fitness function only the ones needed. For example, i have a vector of coefficients W that i want to optimize, but its length is dependent on an integer variable I, ie; If i have I=2 i need a vector W with 2 elements, if I=4 i need W with 4 elements. If the maximum number for I is 10. Then:

    h = @(X) NETWORK_mex(X);
    nvars=11;
    LB=[1 -1*ones(1,10)]
    UB=[10 ones(1,10)]
    [x, err] = ga(h, nvars,[],[],[],[],LB,UB,[],1,ga_opts);

And inside the fitness function:

    function J=NETWORK(X)

     I=X(1);
     W=X(2:I+1);
     ...