🔎NL2SQL360

💫Overview

NL2SQL360 is a testbed for fine-grained evaluation of NL2SQL solutions. Our testbed integrates existing NL2SQL benchmarks, a repository of NL2SQL models, and various evaluation metrics, which aims to provide an intuitive and user-friendly platform to enable both standard and customized performance evaluations. Users can utilize NL2SQL360 to assess different NL2SQL methods against established benchmarks or tailor their evaluations based on specific criteria. This flexibility allows for testing solutions in specific data domains or analyzing performance on different characteristics of SQL queries.

In addition, we propose SuperSQL, which achieves competitive performance with execution accuracy of 87% and 62.66% on the Spider and BIRD test sets, respectively.

🎉News

[24/9/23] We release NL2SQL360 1.1.0 version, which supports two new metrics Reward-based VES (RVES), Soft-F1 Score (F1), from BIRD-Mini-Dev dataset. Please update your package with pip install --upgrade nl2sql360.

[24/9/1] We have released our Homepage & Leaderboard!

[24/8/2] We have released CLI usage / Code usage tutorials. Please check out!

[24/7/30] We have refactored the code and released the official python package(nl2sql360 · PyPI). Stay tuned for the complete documents!

[24/6/30] Our paper The Dawn of Natural Language to SQL: Are We Fully Ready? has been accepted by VLDB'24.

🎈Features

• Easy-to-use Evaluation: Command Line Usage / Python Code Usage.
• Integrated Metrics: Execution Accuracy / Exact-Match Accuracy / Valid Efficiency Score / Question Variance Testing.
• Multi-angle Performance: Fine-grained performance (JOIN, Sub-query, etc.) / Scenario-based (Business Intelligence, etc.)

🔧Installation

pip install nl2sql360

🚀Quick Start

Prepare Dataset

Download NL2SQL dataset to DATASET_DIR_PATH. The directory structure should be like:

DATASET_DIR_PATH:
├─database
│  ├─academic
│  │  ├─academic.sqlite
│  ├─college
│  │  ├─college.sqlite
├─dev.json
├─tables.json

• database directory contains multiple subdirectories, which include the corresponding sqlite database file.
• dev.json is the samples file in JSON format, which at least contains three keys for NL Question, Gold SQL, Databae Id. You can also add the key for Sample Complexity for categorizing samples into different difficulty levels.
• tables.json contains all database schema, following Spider Preprocess Procedure. You can also ignore this file if you do not want to evaluate Exact-Match Accuracy Metic.
• Note that the name for database directory, samples file dev.json and tables file tables.json can be changed.

Import Dataset into NL2SQL360

• CLI Usage:

  • Create / Modify the YAML configuration following NL2SQL360/examples/cli_examples/dataset_spider.yaml.

  • Save the YAML file to the path DATASET_YAML_PATH. Then run the command line:

    nl2sql360-cli dataset DATASET_YAML_PATH

• Code Usage:

  • Create / Modify Python File following NL2SQL360/examples/py_examples/dataset_import.py.
  • Run the python file to import dataset.

Evaluation NL2SQL Model

• CLI Usage:

  • Create / Modify the YAML configuration following NL2SQL360/examples/cli_examples/evaluation.yaml.

  • Save the YAML file to the path DATASET_YAML_PATH. Then run the command line:

    nl2sql360-cli evaluate DATASET_YAML_PATH

• Code Usage:

  • Create / Modify Python File following NL2SQL360/examples/py_examples/evaluation.py.
  • Run the python file to evaluate the model.

Query Multi-angle Performance

• CLI Usage:

  • Create / Modify the YAML configuration following NL2SQL360/examples/cli_examples/report.yaml.

  • Save the YAML file to the path DATASET_YAML_PATH. Then run the command line:

    nl2sql360-cli report DATASET_YAML_PATH

  • The generated report will be in save_path specified in the YAML file.

• Code Usage:

  • Create / Modify Python File following NL2SQL360/examples/py_examples/report.py.
  • Run the python file to generate report.

Delete History Cache

• CLI Usage:

  • Create / Modify the YAML configuration following NL2SQL360/examples/cli_examples/delete_history.yaml.

  • Save the YAML file to the path DATASET_YAML_PATH. Then run the command line:

    nl2sql360-cli delete DATASET_YAML_PATH

• Code Usage:

  • Create / Modify Python File following NL2SQL360/examples/py_examples/delete_history.py.
  • Run the python file to delete dataset / evaluation cache.

🎯Road Map

✅Release NL2SQL360 evaluation code.

✅Release NL2SQL360 experiments data.

✅Release NL2SQL360 Official Python Package.

💾Experiment Data

We have released all experiment data used in our paper.

Download Link

📌Citation

@misc{li2024dawn,
      title={The Dawn of Natural Language to SQL: Are We Fully Ready?}, 
      author={Boyan Li and Yuyu Luo and Chengliang Chai and Guoliang Li and Nan Tang},
      year={2024},
      eprint={2406.01265},
      archivePrefix={arXiv},
      primaryClass={id='cs.DB' full_name='Databases' is_active=True alt_name=None in_archive='cs' is_general=False description='Covers database management, datamining, and data processing. Roughly includes material in ACM Subject Classes E.2, E.5, H.0, H.2, and J.1.'}
}