Repo dedicated to different approaches of transforming natural language questions to structured ones such as SQL

SQLova

• Wonseok Hwang, Jinyeong Yim, Seunghyun Park, and Minjoon Seo.
• Affiliation: Clova AI Research, NAVER Corp., Seongnam, Korea.
• The updated version of manuscript is available from arXiv.

How to run examples

Requirements

Tested on Python 3.8 and Cuda 11.3

• Install minicoda and following dependencies
  • conda install pytorch torchvision -c pytorch
  • conda install -c conda-forge records
  • conda install babel
  • conda install matplotlib
  • conda install defusedxml
  • conda install tqdm
  • conda install -c conda-forge stanfordcorenlp
  • conda install -c anaconda sqlalchemy
  • conda install -c ericmjl isort
  • conda install -c conda-forge black

Models and data

Download annotated WikiSQL data and the PyTorch-converted pre-trained BERT parameters here and place whole folder under data folder

Run

python3 train.py you will be prompted with topic lookup and then question