How dllarray works in Matlab
8 visualizzazioni (ultimi 30 giorni)
Mostra commenti meno recenti
If I have a data with dimension = 1024x1. When I predict the result using a trained network net which takes input 1(C)x1(B)x2048(T) and after that when I check the mse from two different method why do they generate two different answers?
A = dlarray(data,'TCB')
B = predict(net,A);
loss = mse(A,B)
loss = 28.6
loss = mse(squeeze(extractdata(B)),squeeze(extractdata(A)));
loss = 0.026
I need to work on dlarray for autodifferntiation. Please someone guide me
4 Commenti
Risposte (1)
Sahas
il 9 Set 2024
In the first method mentioned, MATLAB’s “mse” function takes two formatted “dlarray” objects as input and computes the MSE. But in the second method, the “mse” function takes two standard datatypes, which are the underlying datatypes of the “dlarrray” objects and computes the MSE.
The MathWorks documentation for “mse” function states that the first input argument, “prediction”, must be a formatted or unformatted “dlarray” object. When “extractdata” function is used in method 2, the output datatype is the underlying datatype of the input “dlarray” object which results in an incorrect MSE.
Refer to the following MathWorks documentation for more information on input arguments of the “mse” function: https://www.mathworks.com/help/deeplearning/ref/dlarray.mse.html
I suggest using the first method to calculate MSE. Refer to the following code snippet for reference:
clc
A = dlarray(data, 'TCB')
B = predict(net,A)
loss = mse(A,B) %Correct usage
% A2 = squeeze(extractdata(A))
% B2 = squeeze(extractdata(B))
Atemp = extractdata(A)
Btemp = extractdata(B)
losstemp = mse(Btemp, Atemp) %Incorrect usage
% losstemp = mse(B, Atemp) %Correct usage
A2 = squeeze(A)
B2 = squeeze(B)
loss2 = mse(A2, B2) %Alternate correct use
For more information on the usage of “extractdata” function and data formats of “dlarray” objects, refer to the following MathWorks documentation links:
- https://in.mathworks.com/help/deeplearning/ref/dlarray.extractdata.html
- https://www.mathworks.com/help/deeplearning/ug/deep-learning-data-formats.html
Hope this is beneficial!
0 Commenti
Vedere anche
Categorie
Scopri di più su Parallel and Cloud in Help Center e File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!