The code here implements the one described in the paper Key-Value Retrieval Networks for Task-Oriented Dialogue https://arxiv.org/pdf/1705.05414.pdf

Dataset Thanks to the authors for making the dataset public. https://nlp.stanford.edu/blog/a-new-multi-turn-multi-domain-task-oriented-dialogue-dataset/

Install the dependencies

numpy==1.14.5
pandas==0.23.3
pkg-resources==0.0.0
python-dateutil==2.7.3
pytz==2018.5
six==1.11.0
torch==0.4.0

To Run the training use

python run.py

To run the examples once training is done

python examples.py

To run Bleu evaluation

cd utils/
python bleutest.py

We get Bleu score of 0.43 which is same as original Keras code accuracy at https://github.com/sunnysai12345/KVMemnn/blob/master/utils/Bleu.ipynb

MIT