search

XMUNLP / XMUNMT

An implementation of RNNsearch using TensorFlow
BSD 3-Clause "New" or "Revised" License
68 stars 24 forks source link

[Question] Does dropout layer after word embedding improve neural machine translation? #12

Open SkyAndCloud opened 6 years ago

SkyAndCloud commented 6 years ago

Hi, I saw you add dropout layer after word embedding, which was not mentioned in rnnsearch paper "Neural Machine Translation by Jointly Learning to Align and Translate". Does this trick improve some performance? Is this implemented in vanilla theano version groundhog? Thanks!

Playinf commented 6 years ago

We found this trick works well in practice. However, the original GroundHog implementation do not use this trick.