This the code of the paper 'Domain-Lifelong Learning for Dialogue State Tracking via Knowledge Preservation Networks'. Python=3.8.5 torch=1.5.0 transformers=3.2.0 numpy=1.19.4
Steps:
create_data.py
: downloads MultiWOZ and splits it by train/valid/testcreate_lifelong_data21.py
: split data by domain for continual learning set uptrain_test.py
: run training for Knowledge Preservation Network (KPN)
- ignore the MultiWOZ
.rar
file. just runningcreate_data.py
will download data and extract it.