search

StefOe / indrnn-pytorch

pytorch implementation of Independently Recurrent Neural Networks https://arxiv.org/abs/1803.04831
121 stars 33 forks source link

Dropout #5

Open sbrugman opened 6 years ago

sbrugman commented 6 years ago

The paper mentions

"Dropout [9] was applied after each IndRNN layer with a dropping probability of 0.25 and 0.1 for CS and CV settings, respectively."

and

"Dropout [9] with a dropping probability of 0.35 were used among IndRNN layers (including embedding) while 0.8 is used after the last IndRNN layer."

It would be nice to have that included too, i.e. https://pytorch.org/docs/master/_modules/torch/nn/modules/dropout.html

LiyangLiu123 commented 5 years ago

I am redoing the thesis on NTU skeleton dataset. I'm also so confused about where should the dropout be put because I didn't see it in the code. When I redo the code, I just put dropout at the end of the loop of the forward function in model indrnn. But the result is not so good