My MSRA projects of implementations of papers.
A chatbot based on seq2seq model, implemented by Theano, based on the Google paper A Neural Conversational Model.
The training dataset can be found here: Cornell Movie--Dialogs Corpus
========
First, have theano installed.
Download the dataset above, make a new folder called data for it.
To prepare the data for training, run make_convs.py.
Next run build_model.py to train:
$THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python build_model.pyNote that the last few lines are commented out, which illustrate how to make a Chatbot instance with a model instance, and to talk to the chatbot instance by simply passing it a string.
========