Passing class method to fminunc

Hello, comrades!

I've ran into a problem: when I use fminunc(), it requires the first argument to be the handle of the investigated function. Everything was fine, when I made separate .m-file with my function. However, the project has grown up since that time, and now this function is a method of a class, and I need to pass it to fminunc (it is situated in another method, see sample code)

classdef MyProblem < handle
    properties 
        N    
        ...
    end
        
    methods
        <other methods>
            
            
         function f = F(obj, par)
            <function body>  
            %%% par is a 1 x obj.N array of numbers
            %%% this method returns one number
        end
            
        function [x,fval,exitflag,output,grad,hessian] = optim_fminunc(obj, x0)
            %% This is an auto generated MATLAB file from Optimization Tool.
            %% Start with the default options
            options = optimoptions('fminunc');
            %% Modify options setting
            %% Here I modify options for the optimization
            
            
            [x,fval,exitflag,output,grad,hessian] = ...
            fminunc(@(x) F(obj, x),x0,options);
        end
        end 
            
    end
end

Here I get an error "Dot indexing is not supported for variables of this type."

Is it even possible to optimize function F from such a class? What should I do if it is so?

Thank you in advance!

P.S. If my code is not ok, and it is strongly recommended to place here a sample that could be ran, let me know.

Answer 1

[Raag]

Hi Anton,

To resolve the error, you need to update the function handle so that the method is bound to the object instance. Instead of writing:

fminunc(@(x) F(obj, x), x0, options);

resolve

fminunc(@(x) obj.F(x), x0, options);

Here is a code that illustrates the change in context:

classdef MyProblem < handle 

   properties 

       N  % Number of variables 

   end 

   methods 

       % Constructor to set the number of variables 

       function obj = MyProblem(N) 

           obj.N = N; 

       end 

       % Objective function (for example, the sum of squares) 

       function f = F(obj, x) 

           f = sum(x.^2); 

       end 

       % Optimization method using fminunc 

       function [x, fval] = runOptimization(obj, x0) 

           options = optimoptions('fminunc', 'Display', 'iter'); 

           % Bind the function handle to the object using dot notation 

           [x, fval] = fminunc(@(x) obj.F(x), x0, options); 

       end 

   end 

end 

By using ‘@(x) obj.F(x)’, MATLAB correctly recognizes ‘F’ as a method of ‘obj’, which avoids the "Dot indexing is not supported" error.

For a better understanding of the above solution, refer to the following MATLAB documentation:

1. fminunc: https://www.mathworks.com/help/releases/R2022a/optim/ug/fminunc.html