Pytorch RNN with no nonlinearity

Is it possible to implement an RNN layer with no nonlinearity in Pytorch like in Keras where one can set the activation to linear? By removing the nonlinearlity, I want to implement a first-order infinite-impulse-response (IIR) filter with a differentiable parameter and integrate it into my model for end-to-end learning. I can obviously implement the filter in Pytorch but I thought using an inbuilt function may be more efficient.

https://stackoverflow.com/questions/66726974/pytorch-rnn-with-no-nonlinearity March 21, 2021 at 06:13AM