LION-Net: LIghtweight ONtology-independent Networks for Schema-Guided Dialogue State Generation
This repo contains source code of our DSTC8-trackIV 2020 paper "LION-Net: LIghtweight ONtology-independent Networks for Schema-Guided Dialogue State Generation"
https://drive.google.com/file/d/1d0S_i0eLYUBpTgs_AoyU0L106iI5WB3w/view?usp=sharing
Requirements
- Python >= 3.6
Required python packages are listed in requirements.txt.
Dataset
The dataset we used is Schema-Guided Dialogue State Tracking Dataset provided by Google.
https://github.com/google-research-datasets/dstc8-schema-guided-dialogue
Preprocess
Download the dataset first and remember to download the GloVe word vectors.
After downloading you need to put them into the directory you want.
Create dataset
python3 preprocess.py
python3 extract_schema.py
Training
First, you need to make a copy of config.yaml.example and change the name to config.yaml Then you can change the parameters in config.yaml and do the training. Sample usage:
python3 train.py
Testing
Sample usage:
python3 test.py