How to save a variable from an anonymous function?

I am using the function fmincon to minimize the function f:

[X,f]=fmincon(@(X) obj_function(X,P,Pd,intervalotiempo),X0,A,B,Aeq,Beq,LB,UB,@(X)nonlcon(X,Potnominal,intervalotiempo,Pd));

fmincon requires the use of anonymous funtions, therefore the variables that are used in nonlcon are not stored after runing the program.

I would like to know if it is possible to save (get it as an output) one varible from the function nonlcon.

Answer 1

[Walter Roberson]

There is no mechanism for returning values from nonlcon.

You can any of the usual techniques for saving data from a function. For example you could make nonlcon a nested function with a shared variable.

Note that the objective function may be invoked multiple times without a nonlcon invocation, and that nonlcon may get invoked multiple times without invoking the objective function, including at locations that the objective function is not called for. You should not count on the calls to the two functions corresponding.

Answer 2

[Stephan]

for nonlcon it's not necessarily an anonymous function that is needed. you can also use a "normal" function, so that you can save the needed values or display them.

[Matt J]

In fact, I cannot think of any Matlab command that cares which way an input function handle is implemented (except maybe func2str).

[Maria Sanz]

And how can I rewrite: @(X)nonlcon(X,Potnominal,intervalotiempo,Pd) in a way that an anonymous function is not needed?

I need that X varies during the execution of fmincon.

[Walter Roberson]

function driver
   various things here
   Potnominal = value;
   intervalotiempo = value;
   Pd = value;
   
   stored_intermediate_results = {};
   
   [x, fval] = fmincon(@objective_function, A, b, Aeq, beq, lb, ub, @nonlcon, options);
   
   save('AppropriateName.mat', 'x', 'fval', 'stored_intermediate_results');
   
    function [c, ceq] = nonlcon(X)
        %here Potnominal, intervalotiempo, Pd, stored_intermediate_results are all shared variables
 
        temp1 = some expression in X and Potnominal and intervalotiempo and Pd
        temp2 = some expression in X and Potnominal and intervalotiempo and Pd
        c = some expression in X and Potnominal and intervalotiempo and Pd and temp1 and temp2
        ceq = some expression in X and Potnominal and intervalotiempo and Pd and temp1 and temp2
        stored_intermediate_results(end+1) = {X, temp1, temp2};
    end
end

No anonymous function. But it would not have mattered, because you could instead use

function driver
   various things here
   Potnominal = value;
   intervalotiempo = value;
   Pd = value;
   
   stored_intermediate_results = {};
   
   [x, fval] = fmincon(@objective_function, A, b, Aeq, beq, lb, ub, @(X)nonlcon(X,Potnominal,intervalotiempo,Pd), options);
   
   save('AppropriateName.mat', 'x', 'fval', 'stored_intermediate_results');
   
    function [c, ceq] = nonlcon(X,Potnominal,intervalotiempo,Pd)
        %here Potnominal, intervalotiempo, Pd are all parameters
        %here stored_intermediate_results is a shared variable

        temp1 = some expression in X and Potnominal and intervalotiempo and Pd
        temp2 = some expression in X and Potnominal and intervalotiempo and Pd
        c = some expression in X and Potnominal and intervalotiempo and Pd and temp1 and temp2
        ceq = some expression in X and Potnominal and intervalotiempo and Pd and temp1 and temp2
        stored_intermediate_results(end+1) = {X, temp1, temp2};
    end
end

There is a small execution time difference between these two, with the parameter version being slightly faster than the shared variable version.