deep-learning-from-scratch-3 icon indicating copy to clipboard operation
deep-learning-from-scratch-3 copied to clipboard

Published 20 hours ago • verified publisher icon oreilly-japan

Fix: RNN layer's h2h should be in_size=hidden_size

Open shionhonda opened this issue 4 years ago • 0 comments

RNN.h2h should be initialized with in_size=hidden_size.

steps/step59.py is currently working because it instantiates RNN with L.RNN(hidden_size) (i.e., in_size=None), but if you instantiate like L.RNN(hidden_size, 1), it throws a shape mismatch error at self.h2h(self.h) during its forward operation.

shionhonda avatar Aug 26 '21 00:08 shionhonda