mkolod / mxnet_seq2seq Goto Github PK

In get_s2s_data the training files are scanned to find the vocabulary, and only after that they are tokenized, using the vocabulary for converting the tokens into indeces.

I think the order should be -> tokenization -> vocabulary -> convert into indeces

I think it is strange that we need to remove the state feeding when using the attention.

The other NMT systems initialize the decoder with something coming from the encoder.
For example:
Nematus initialize the decoder with the mean of the encoder outputs
https://github.com/rsennrich/nematus/blob/73037e94884fd2d1c1d18d81686cd1f6ea32d073/nematus/nmt.py#L421-L440

While OpenNMT is passing to the decoder the states returned from the unrolling of the encoder
https://github.com/mattiadg/OpenNMT-py/blob/6bf832170421ca9c3b609ab3cfea22ae5b7fdb4e/onmt/Models.py#L140-L146

At the moment I don't understand the OpenNMT code because I don't know pytorch, but I'll try to understand what they are doing and implement it here.