How to run the code?

[RisabBiswas]

Hey! Great work.

Can you add steps to run/experiment with the code base?

Thanks!

[takan55]

+1 for this

[Satissss]

Here are the steps to obtain the schema for SQL generation using LinkAlign:

1. Preprocess the data:

python preprocess.py --base_dir ./spider2_dev

2. Generate the schemas:

python generate_schemas.py \
    --save_path ./spider2_dev/instance_schemas \
    --schema_path ./spider2_dev/schemas \
    --dataset ./spider2_dev/spider2_dev_preprocessed.json \
    --db_info_path ./spider2_dev/db_info.json \
    --external_info_path ./spider2_dev/external_knowledge

By the way, you can leverage components from tools/SchemaLinkingTool.py (like in generate_schemas.py) to build more effective schema linking pipelines.

[JaamieZheng]

How can I get the similar preprocessed data of Spider2.0-snow ?

[aaafei123]

How can I get the similar preprocessed data of Spider2.0-snow ?

[Satissss]

Hi, we’ve developed Squrve, a lightweight, config-driven Text-to-SQL framework that decouples methods from dataset formats. It supports benchmarks like Spider, BIRD, Spider 2.0 (lite, snow), handling structure, preprocessing, and evaluation automatically. You can simply set datasource to the dataset identifier (e.g., Spider2:lite:) to test and compare their performance.