Open z-zawhtet-a opened 8 years ago
I will try: https://github.com/farizrahman4u/seq2seq for the character prediction
The tensorflow sequence to sequence model runs at 800 words per second on my 24 cores computer. If keras can do any better, then lets switch to it!
Keras won't do much better since we use tensorflow backend. Can you try running it on chunk of reviews?
can we change the backend? I think tensorflow is too slow
Yeah, we should try sequence to sequence prediction using Keras. It's much simpler, great way to test our idea.