Update tutorial #3 to match new TF seq2seq API
ematvey opened this issue · comments
Whenever it is released.
it seems that the seq2seq API changes very frequently. I tried to run tutorial #3 under tensorflow 1.0.0 and it didn't run, but 1.0.0-alpha1 worked for me a couple of days ago.
it seems that the seq2seq API changes very frequently.
It's true.
At the time I wrote this, tutorial 3 works on pip-installed tensorflow, but doesn't work on compiled-from-source tensorflow. (both latest and show version 1.0.1)
However, Google's seq2seq repo works on both. (at least for the rnn_encoder_test)
After they stabilize API a bit I'll update tutorial 3. It also might not be necessary if they release their own sufficienctly clear tutorials, althout at this time they went with YAML configs instead of code examples (https://google.github.io/seq2seq/nmt/).
Many thanks to the pytorch seq2seq example! Seems like pytorch codes are more friendly to beginners.