# Conducting automatic differentiation on simple neural network (.net) file

6 visualizaciones (últimos 30 días)
Tomey el 29 de Mzo. de 2023
Comentada: Tomey el 13 de Abr. de 2023
Hi,
I would like to conduct an automatic differentiation on a simple neural network. I know this can be done on deep neural networks. I already tried this but I receive an error message saying 'value to differentiate must be a traced dlarray scalar'.
Here is my code
function [Y, dydA,dydB, dydC, dydD] = Y_partial(A,B,C, D)
[dydA,dydB, dydC, dydD] = dlgradient(Y, A, B,C, D);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
A = dlarray(A);
B = dlarray(B);
C= dlarray(C);
D = dlarray(D);
[Y,dydA,dydB, dydC, dydD] = dlfeval(@Y,A,B,C, D);
##### 0 comentariosMostrar -2 comentarios más antiguosOcultar -2 comentarios más antiguos

Iniciar sesión para comentar.

### Respuestas (1)

Abhijeet el 3 de Abr. de 2023
Hi,
The error message indicates that the input to the dlgradient must be a scalar value, but Y in your code seems to be a loaded neural network. To use dlgradient, you need to perform a forward pass through the neural network and obtain a scalar output that you want to differentiate with respect to the inputs.
function [Y, dydA, dydB, dydC, dydD] = Y_partial(A, B, C, D)
Y = predict(net.net, A, B, C, D); % Performing a forward pass
[dydA, dydB, dydC, dydD] = dlgradient(Y, A, B, C, D);
end
A = dlarray(A);
B = dlarray(B);
C = dlarray(C);
D = dlarray(D);
[Y, dydA, dydB, dydC, dydD] = dlfeval(@Y_partial, A, B, C, D);
Here, predict performs a forward pass through the neural network to obtain the scalar output Y, which can then be used as the first argument to dlgradient. Y_partial is modified to take the input values as arguments and perform the necessary computation to produce Y and the gradients. After this, dlfeval is used to evaluate Y_partial with the input arguments specified as dlarray objects.
Thanks
##### 3 comentariosMostrar 1 comentario más antiguoOcultar 1 comentario más antiguo
Abhijeet el 5 de Abr. de 2023
Welcome Tomey !!
Yes, Sim should work and resolve the issue that you're facing. You can still dig into the documentation of sim and predict to explore more using the commands.
help sim, doc sim, help predict, doc predict
Thanks Again !!
Tomey el 13 de Abr. de 2023
Thank you. Abjijeet. It seems that the sim function does not take dlarrays number which is applicable to only the automatic differentiation (dlgradient). I tried to create a new variables that can be fed into the sim function and then use the dlarray variables on a the dlgradient. But it doesnt seem to work. What do you suggest?

Iniciar sesión para comentar.

### Categorías

Más información sobre Custom Training Loops en Help Center y File Exchange.

### Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by