This is my first attempt on a simple project to create a language translation model using Sequence To Sequence Learning Approach.
Details can be found at my blog post here:
I used the Europarl's Parallel Corpus for training. To get the source code to work immediately, you have to use the newest version (release v8 at the time of writing) at the link below (following the link will start a 180MB download):
Feel free to change the default dataset to anyone of your own. Just don't forget to modify the code!
python seq2seq.py
# Max length:= 300, number of recurrent layers:= 2, dimension of hidden state:= 500
python seq2seq.py -max_len 300 -layer_num 2 -hidden_dim 500
The network must be trained at least once (trained weights must exist!).
python seq2seq.py -mode test
# Max length:= 300, number of recurrent layers:= 2, dimension of hidden state:= 500
python seq2seq.py -mode test -max_len 300 -layer_num 2 -hidden_dim 500