Custom cells for RNNs in PyTorch (trying to implement EntNet)


#1

Hi all,
New to PyTorch, I’d like to implement the EntNet model, as I’ve seen only TensorFlow versions, not PyTorch.

The tf implementations (for example here) seem to use tf.nn.dynamic_rnn with a custom dynamic memory cell for the cell argument. Is there such an equivalent in PyTorch? I haven’t found one.

The original author’s Torch implementation does not use the custom cell approach, but there is something elegant about it, so I thought to follow it if possible.

If you have any other recommendations, I’d be happy to hear them also :slight_smile:

Thanks!