can i implement lstm layer from scratch using matlab in sequence to sequence regression

1)can i implement lstm layer from scratch instead of pretrained network used in this example?

Sequence-to-Sequence Regression Using Deep Learning - MATLAB & Simulink (mathworks.com).

2)can i change the activation functions for all cells in lstm layer?

Answer 1

[Ben]

Sure, I think the best way to do this would be with a custom layer, you can follow this example that implements a modified LSTM called peephole LSTM using custom layers:

https://uk.mathworks.com/help/deeplearning/ug/define-custom-recurrent-deep-learning-layer.html

[walaa gomaa]

thanks,alot.

i will try it