Error: Error using dlfeval Value to differentiate must be a traced dlarray scalar.

Hello, I am working on auto differentiation. But it came up with a error shown as title.

Error: Error using dlfeval Value to differentiate must be a traced dlarray scalar.

I am not sure why this doesn't work. The benchmark function is CEC2005 benchmark function.

x=bnd(1)+(bnd(2)-bnd(1)).*rand(1,dim); % it's just random assinged x, can also be replaced by [0,0]
x0 = dlarray(x); % wrap it with dlarray
[fval,g] = dlfeval(@modifiedBenchmark,x0);

function [f,grad] = modifiedBenchmark(x)
    f = @(x)benchmark_func(x,1); % 1 here means that I use the first benchmark function
    grad = dlgradient(f,x);    
end

I also tried the sample code on the official website.

% Example code
x0 = dlarray([-1,2]);
[fval,gradval] = dlfeval(@rosenbrock,x0)

function [f,grad] = rosenbrock(x)
f = 100*(x(2) - x(1).^2).^2 + (1 - x(1)).^2;
grad = dlgradient(f,x);
end

And it works fine.

Can anyone show me what's going wrong?: (

Thank you!

[Walter Roberson]

Related: https://www.mathworks.com/matlabcentral/answers/717068-dlgradient-error-value-to-differentiate-must-be-a-traced-dlarray-scalar#answer_600275

https://www.mathworks.com/matlabcentral/answers/493008-what-is-derivative-trace-in-dlgradient-function#answer_410106

[Chiao-Ling Cheng]

Hello Walter,

Thank you for the links. But I still can't find the problem here after reading these two posts. Do you have any ideas? Thank you!

Answer 1

[Walter Roberson]

Notice the example code has

f = 100*(x(2) - x(1).^2).^2 + (1 - x(1)).^2;
grad = dlgradient(f,x);

so it is receiving an x (that is dlarray object), computing something with it, getting out a dlarray object, and passing that dlarray object to dlgradient along with the original dlarray that was input.

Whereas your attempted code has

    f = @(x)benchmark_func(x,1); % 1 here means that I use the first benchmark function
    grad = dlgradient(f,x);    

Which constructs an anonymous function and passes the anonymous function to dlgradient. But dlgradient does not accept function handles.

You would need something like

    f = benchmark_func(x,1); % 1 here means that I use the first benchmark function
    grad = dlgradient(f,x);