In the process of text coding encoder can separate training? Or say only and decoder joint training?
Also at the time of verification model if the input word is not in the training of the word list?
Of new just into the pit is not very good