Closed jeff1evesque closed 5 years ago
Rather than reinventing an existing implementation, we can leverage an existing train.py
, which also makes use of the nmt
neural network methodology.
Given limited resources, and time, creating a chatbot ground up using sequential neural network is not practical. Instead, we'll create a chatbot using existing code, then create front-proxy. This will likely be a simple classifier rule that can simply predict if a statement is a question. Then, the chatbot will focus on the user submitting a question of interest. Therefore, for this issue, we'll eliminate previous nmt code, and simply version control the finished chatbot dropbox link (chatbot will exceed 100MB github limit).
We need to implement tensorflow with sequence to sequence training.