search

oswaldoludwig / Seq2seq-Chatbot-for-Keras

This repository contains a new generative model of chatbot based on seq2seq modeling.
Apache License 2.0
332 stars 98 forks source link

Not a big deal, but Ubuntu does not like the spaces in the names #22

Open BlogBlocks opened 6 years ago

BlogBlocks commented 6 years ago

Not a big deal, but Ubuntu 16.04 Python 2.7 0 does not like the spaces in the names. Just replace the spaces with a - and all works well. Example: name='the-context-text name='the-answer-text-up-to-the-current-token Do all the names that way.

Question: Does the bot auto-learn from conversation or must it be retrained with each session ?

oswaldoludwig commented 6 years ago

The bot saves the conversation for further training, see lines 124-126, 188 and 189 of conversation.py. Therefore, you can "fine tune" the pre-trained model using these files to train it for a few epochs.

oswaldoludwig commented 6 years ago

It is worth mentioning here that the script conversation_discriminator.py performs better than conversation.py.

BlogBlocks commented 6 years ago

Shouldn't the files (file_saved_context and file_saved_answer) in conversation.py be open to append? qf = open(file_saved_context, 'a') af = open(file_saved_answer, 'a') That way you can keep adding to the files until training is initiated ?

oswaldoludwig commented 6 years ago

Yes, it can be that way too. Either way, the idea is to edit these files to select the best utterances before using it for training.