Spider is a large-scale semantic parsing and text-to-SQL dataset. We parse schema SQL files from the development set and retrieve join paths between primary keys and foreign keys as ground truth.
data.zip contains raw tables in the CSV format.
dev_metadata.csv annotates the ground truth of join paths.
data_loader.py provides functionalities to load tables and metadata.
If you find this dataset useful, please consider citing the following paper.
@inproceedings{DBLP:conf/cidr/CongGFJD23,
author = {Tianji Cong and
James Gale and
Jason Frantz and
H. V. Jagadish and
{\c{C}}agatay Demiralp},
title = {WarpGate: {A} Semantic Join Discovery System for Cloud Data Warehouses},
booktitle = {13th Conference on Innovative Data Systems Research, {CIDR} 2023, Amsterdam, The Netherlands, January 8-11, 2023},
publisher = {www.cidrdb.org},
year = {2023}
}