Closed christopher5106 closed 7 years ago
True, good catch. I'll send a fix. Thanks!
By the way, do you get good convergence results on Theano ? Thanks
Hi Christopher, Yes, got the same results as in paper for PTB (there's a log in #2 ). Thanks for pointing this out, please see #6 .
Hi Julian, When you train over multiple layers https://github.com/julian121266/RecurrentHighwayNetworks/blob/master/theano_rhn.py#L225 I'm not sure you really reset the hidden state, since the reset method will correspond to the last layer only. Am I wrong ?