Error using makeState (line 56) Your fitness function must return a scalar value.

Hello Matlab users

I'm getting started with GA and trying to apply it to minimization problem I have. Running the Genetic Algorithm gives the following error:

Error using makeState (line 56)
Your fitness function must return a scalar value.

Error in galincon (line 17)
state = makeState(GenomeLength,FitnessFcn,Iterate,output.problemtype,options);

Error in ga (line 359)
            [x,fval,exitFlag,output,population,scores] = galincon(FitnessFcn,nvars, ...

Error in minimizer (line 45)
[x,f] = ga(objfunc, nvars, [], [] , [], [], LB, UB, [],options);

I read somewhere that should run GA function with the following options structure:

options = gaoptimset('Vectorized','off');

I did it but still have this error. I set 'dbstop if error' in my file and after running GA this message is received:

error(message('globaloptim:makestate:fitnessCheck'));

I would be grateful if you help me.

Thanks

Nasrin Sadeghi

[Fahad]

same problem

Answer 1

[Walter Roberson]

Your objfunc needs to be a function handle for a function that returns a single value (a scalar). Currently it is returning a vector or an array.

We would need to see the code for your objfunc in order to say why your code is returning a vector or array.

[nasrin]

Thanks for your answer

function f = evaluator(x)

x1 = [0.00683 0.00924 0.01252 0.0158 0.02093 0.02664 0.03505 0.04089 0.0525 0.06271 0.07792 0.10226 0.15405 0.20271 0.25451 0.30708 0.33655 0.38735 0.43815 0.48895 0.53975 0.59055];

i0 = length(x1);

x2 = [0.0127 0.01905 0.0254 0.03175 0.0381];

 j0 = length(x2);

x3 = 0.05 : 0.01 : 0.5;

k0 = length(x3);

for i = 1 : i0

    for j = 1 : j0

        for k = 1 : k0

             Nt(i,j) = K1 * (x1(1,i) /x2(1,j))^ n1;

            v_t(i,j) = (m_t * n) / (0.785 * (0.8*x2(1,j))^2*density_t*Nt(i,j));
and etc
 f1(i,j,k) = 8000 + (259.2 * A(i,j,k)^0.91);

 f2(i,j,k) = 0;
            for a =1:10
                f2 (i,j,k)= f2(i,j,k) + (1.05/1.1^a) * ((m_t *      P_t(i,j,k) / density_t) + (m_s * P_s(i,j,k) / density_s));
            end
f(i,j,k) = f1(i,j,k) + f2(i,j,k);
        end
    end
end

[nasrin]

function [x, f] = minimizer

objfunc = @(x)evaluator(x);

nvars = 3;

LB = [1 1 1];

UB = [22 5 46];

options = gaoptimset('InitialPopulation',[],'PopulationSize',100,... 'Generations',50,'StallGenLimit',100,'SelectionFcn',@selectionroulette,'FitnessScalingFcn',@fitscalingrank,'Vectorized','off'); [x,f] = ga(objfunc, nvars, [], [] , [], [], LB, UB, [],options);

[Walter Roberson]

In your function, you ignore the input x, and you construct f as a 22 x 5 x 46.

You need to be taking a row vector of values as input and you need to construct from that a scalar output.

The exception to that would be if you set Vectorized 'on' in the options. In that case your objective function should take as input a 2D array which has one row for each member of the population, and you would need to return a scalar value for each.

Your code is using a number of variables that are not parameters, including m_t, n, a, A, and density_t . With the summation it is doing to create f2(), I do not see any obvious way of converting it into a function that creates scalars. It looks to me as if your function is intended to sweep x1, x2, and x3 across a range of values. If so then x1, x2, and x3 should be variables passed within the parameter x, as in

x1 = x(1);
x2 = x(2);
x3 = x(3);

and you should be constructing a scalar result for that particular combination. The range of values to use for each one would then become the lower and upper bounds for those particular elements of x.