How to train "timedelaynet" using "trainNetwork" function in Deep Learning Toolbox?

I have build a Time delay neural network and I wanna avoid use "train" function, and I want to train it using "trainNetwork" function to be able to use 'Adam' optimizer. is it possible ? If yes, How?

Answer 1

[Varun Sai Alaparthi]

Hello Abdelwahab,

As of now it is not possible to train ‘timedelaynet’ using ‘trainNetwork’. If you are interested in using ‘trainNetwork’, it might be better to model your data using ‘LSTM’ or ‘GRU’ networks which are supported by ‘trainNetwork’.

Please refer to this link for examples on time series analysis using our deep learning API:

• Deep Learning with Time Series and Sequence Data — Examples (mathworks.com)

Other possible work around to use ‘trainNetwork’ would be to design custom training loop for ‘timedelaynet’ and use ‘Adam’ optimiser in the training.

Please refer to this link for example on custom training loop:

• Train Network Using Custom Training Loop - MATLAB & Simulink (mathworks.com)