Dzmitry Bahdanau
Dzmitry Bahdanau
Using large vocabularies in NMT is an open problem, there are some papers on that, see e.g. http://arxiv.org/abs/1412.2007 On 13 April 2016 at 08:37, Frédéric Bastien [email protected] wrote: > Can...
I think @dwf is right that a centralized server is utopia. A server for a project sounds more realistic.
Agreed, nice idea, adding CCW.
+1 for a folder with README for each experiment Another thing why big files do not cut it: for models to be unpicklable, nothing should be in `__main__` namespace. On...
Honestly I do not see that would be a big improvement over the `__init__.py`/`__main__.py` breakdown we have now.
@orhanf , would you mind? It we go this way, you will not have to wait for reviews for weeks to make changes in your code. (don't pay attention to...
@orhanf , thanks for your quick response and active position. Regarding you first concern: indeed, every example will have to be regularly synchronized to be compatible with the latest Blocks....
Hey guys, this is just to say that I agree that the MT example should be in separate repository. reverse_words was intended to be the toy example of Encoder-Decoder with...
Thanks for raising this issue, @kudkudak For now, here is a brain dump of what I think about it. First of all, people who use `continue_training` won't be affected. It's...
While the canonical way two "switch off" a regularization in Blocks is definitely to keep the original graph, I think that all regularization parameters should be shared variables, properly tagged...