if the answer is very long

[zhoucrawl]

For this model, if the answer is very long, the answer is not good。
In chat.py, the answer is up to 15 words.
How to modify the program?

[uwittygit]

HI there, I run python train.py directly and stop the train at step = 3900, the chat result is awful,
what was your test.py outputs look like?

chengiant@127:~/workspace/tensorflow/work/chatbot$ python test.py
Reading model parameters from /var/lib/tf_seq2seq_chatbot/nn_models/model.ckpt-3900
('Hi!', ' -> ', "+++$+++ i ' m sorry .")
('Hi', ' -> ', '')
('what ?', ' -> ', "i ' m sorry .")
('why ?', ' -> ', "i ' m not a _UNK .")
('who ?', ' -> ', "+++$+++ i ' m not a _UNK .")
('yeah ?', ' -> ', 'yeah .')
('what is it ?', ' -> ', "i ' m sorry .")
('why not ?', ' -> ', "i ' m not a little .")
('really ?', ' -> ', 'yes .')
('huh ?', ' -> ', 'yes .')
('yes ?', ' -> ', "+++$+++ i ' m not going to be .")
("what ' s that ?", ' -> ', "i ' m sorry .")
('what are you doing ?', ' -> ', "i ' m sorry .")
('what are you talking about ?', ' -> ', "i ' m sorry .")
('what happened ?', ' -> ', "i ' m sorry .")
('hello ?', ' -> ', 'yes .')
('where ?', ' -> ', "+++$+++ i ' m not a _UNK .")
('how ?', ' -> ', "+++$+++ i ' m sorry .")
('excuse me ?', ' -> ', 'yes .')
('who are you ?', ' -> ', "i ' m not a _UNK .")
('what do you want ?', ' -> ', "i ' m sorry .")
("what ' s wrong ?", ' -> ', "i ' m sorry .")
('so ?', ' -> ', 'yes .')

[zhoucrawl]

I run python train.py directly and stop the train at step = 17850, perplexity = 14.74.
You can continue to train.

[uwittygit]

@zhoucrawl thanks for your information. I'll take a try

[7oranger]

@zhoucrawl seems you've got a nice result. Maybe I stop the programme too early. I'll try it again.