Total Recall: a Customized Continual Learning Method for Neural Semantic Parsers

This is the code for the EMNLP2021 paper, Total Recall: a Customized Continual Learning Method for Neural Semantic Parsers. This work investigates continual learning for semantic parsing.

@inproceedings{li2021total,
  title={Total Recall: a Customized Continual Learning Method for Neural Semantic Parsers},
  author={Li, Zhuang and Qu, Lizhen and Haffari, Gholamreza},
  booktitle={Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing},
  pages={3816--3831},
  year={2021}
}

Requirements

• Python 3.8
• cuda 11.0

Installation

• Create a new conda environment with python 3.8

conda create -n py38 python=3.8
conda activate py38

• Install pytorch 1.7.1 please select the proper pytorch version given your own environment

pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

• Install other libs

pip install spicy
pip install zss
pip install strsim
pip install quadprog
pip install scikit-learn-extra
pip install nltk
pip install transformers
./pull_dependency.sh

• Run the script to run the best setting in our continual learning experiments

python preprocess_data/nlmap/generate_nlmap_ext_actions.py
./scripts/nlmap/continual_train/city_warm_run/warm_loop_run.sh 0

python preprocess_data/nlmap/generate_nlmap_qtype_actions.py
./scripts/nlmap/continual_train/qtype_warm_run/adap_qtype_loop_run.sh 0

python preprocess_data/overnight/generate_actions.py
./scripts/overnight/continual_train/warm_loop_run.sh 0

• Supervised training on NlMapV2

python preprocess_data/nlmap/generate_nlmap_actions.py
./scripts/nlmap/question_split/train.sh

• Supervised training on NlMapV1

python preprocess_data/nlmap/generate_actions.py
./scripts/nlmap_v1/train.sh