(How) Higher derivative by 'dlgradient' or Higher derivative in matlab

Hi

I am currently coding custom deep learning, but the process stopped at the higher derivative.

I am curious about how to make a higher derivative through dlgradient.

Alternatively, you are welcome to suggest a way to do higher derivatives in matlab.

For example, how to get ddydxx in the following example.

-----------------------------------------------------------------------------------------------------------------------

Code

clc,clear,close all
x=3;
x0=dlarray(x);
[fval,gradval,ggradval] = dlfeval(@Myfunc,x0);
function [fval,gradval,ggradval] = Myfunc(x)
y = 100*(3*x - 7*x.^2).^2; 
dydx=dlgradient(y,x,'RetainData',true);
ddydxx=dlgradient(dydx,x);
end

-----------------------------------------------------------------------------------------------------------------------

Error

Error using dlfeval (line 43)

Value to differentiate must be a traced dlarray scalar.

Error in gradtest (line 7)

[fval,gradval,dd] = dlfeval(@Myfunc,x0,y);

-----------------------------------------------------------------------------------------------------------------------

Thanks for reading my question.

[Walter Roberson]

What is the difference between this and https://www.mathworks.com/matlabcentral/answers/745857-how-higher-derivative-by-dlgradient-or-higher-derivative-in-matlab?s_tid=srchtitle and https://www.mathworks.com/matlabcentral/answers/746647-higher-order-derivative-by-dlgradient?s_tid=srchtitle ?

[jaehong kim]

i am sorry...

I just posted to update the question. Sorry for violating the community rules. I'll be careful.

[Walter Roberson]

https://www.mathworks.com/help/deeplearning/ug/include-automatic-differentiation.html

To use automatic differentiation, you must call dlgradient inside a function and evaluate the function using dlfeval. Represent the point where you take a derivative as a dlarray object, which manages the data structures and enables tracing of evaluation

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

[Maksym Tymchenko]

Hi @jaehong kim,

To make the code above work, you need to set the name value argument "EnableHigherDerivatives" to true.

Here's the modified version of your code that works as expected:

clc,clear,close all
x=3;
x0=dlarray(x);
[fval,gradval,ggradval] = dlfeval(@Myfunc,x0);

function [y,dydx,ddydxx] = Myfunc(x)
y = 100*(3*x - 7*x.^2).^2; 
dydx=dlgradient(y,x,EnableHigherDerivatives=true);
ddydxx=dlgradient(dydx,x);
end