genetic algorithm error not enough input arguments

    function deltaH = curvefit_H_2modes(w1,w2,m1,m2,c1,c2)

    load('file.mat');

    z = -(x.^2) .* ((1 ./ ((w1.^2 * m1) - (x.^2 * m1) + (1i * x * c1))) + ...

        (1 ./ ((w2.^2 * m2) - (x.^2 * m2) + (1i * x * c2))));

    deltaH = sum(abs(z - y).^2);

    end

Ok, now I use the Genetic Algorithm in the Global Optimization Toolbox and input @curvefit_H_2modes as my fitness function and I input 6 as the number of variables. Then I click on Start to run the GA solver. I get an error message saying "Not enough input arguments" ... WHY?! My function takes 6 input arguments. The function loads a file which has two vectors, x and y (experimental data).

I can run the function myself using estimates for the 6 input variables and it runs perfectly. Why does the GA have a problem with it?

[Craig Cowled]

I should add that this is my first time using the Optimization Toolbox, so it's possible I have made a rookie error. Would be grateful if anyone has an answer for me.

Answer 1

[Alan Weiss]

There are several problems with your syntax. The main problem is that all optimization solvers expect the decision variables (the variables you want to change to find an optimum) to be in a single vector, typically called x. For example,

 function deltaH = curvefit_H_2modes(x)
 w1 = x(1);
 w2 = x(2);
 m1 = x(3);
 m2 = x(4);
 c1 = x(5);
 c2 = x(6);
 load('file.mat');
 ...

Furthermore, you will have much better performance if you get the load statement out of your function definition, and simply pass in the values you need (I think it is just y) in a nested function or anonymous function. See the documentation for details.

Finally, you will probably get better results if you use the appropriate solver. For this type of problem (minimizing a sum of squares), the Optimization Decision Table recommends using lsqnonlin or lsqcurvefit. And do you really need to sum the squares, or can you use the norm function?

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

Answer 2

[Craig Cowled]

Alan,

Thank you for taking the time to answer this in so much detail!

Point taken regarding the choice of optimization tool. I think MattJ recommended I use lsqnonlin for this problem as well. Good point about the load command. Will probably slow down the computation. Will look at the norm function.

Actually, I chose the GA because I need to learn how to use this for another problem (updating a finite element model), so I am still very interested in understanding how the function works.

Thank you again, Alan. It's guys like you that make this forum awesome.

Craig.

Answer 3

[Craig Cowled]

Hey, what do you know, it worked!!

Btw, your recommendation of using LSQNONLIN was the right call too.

Although the GA worked, it took a few minutes of computation and the residual was an order of magnitude worse than what I got out of LSQNONLIN. Also, I discovered that I had to tweak some of the GA options to get better results.

If anyone is interested, the fitness function I used for my GA optimization was similar to the code in http://www.mathworks.com.au/matlabcentral/answers/60638-curve-fitting-a-complex-function-using-cftool with the only difference being the last line.

When using LSQNONLIN, I used ...

    deltaH = z1 - y1;

because it outputs a vector which the LSQNONLIN function uses to find the least squares solution.

When using the GA, I changed this to ...

    deltaH = norm(z1 - y1);

because the GA needs a scalar value to calculate the fitness of individuals.

Answer 4

[Ali Meghdadi]

hi, I also have somehow the same problem. My constraint function has 8 parameters as input and gives an output and GA doesn't accept it. what can I do?

[Alan Weiss]

Please look at the accepted answer for the solution to your problem.

If that doesn't work, then provide more detail, such as the syntax you are using, and the error GA returns.

Alan Weiss

MATLAB mathematical toolbox documentation