RYANSQL

Overview

Model description

This model is developed by referring to RYANSQL, a text-to-SQL system for complex, cross-domain databases.

Reference Paper: Choi et al., RYANSQL: Recursively Applying Sketch-based Slot Fillings for Complex Text-to-SQL in Cross-Domain Databases, 2020

The system is submitted to SPIDER leaderboard. The system and its minor improved version RYANSQL v2 is ranked at second and fourth place (as of February 2020).

The system does NOT use any database records, which make it more acceptable to the real world company applications.

Model architecture

input

• inputs are expected to be ELECTRA encoded query and database (schema)
• Shape: (B, 512, 768)

output

• outputs are expected to be ELECTRA
• Shape: (B, 512, 768)

Set up Environment

pip install -f https://download.pytorch.org/whl/torch_stable.html -r requirements.txt

Preprocessing

Download KoNL2SQL Data in 'data/{FOLDER}'

mkdir data/download

Convert KoNL2SQL Data for using model

cd data
python convert_data.py --src_folder download/01.Training/ --name train
python convert_data.py --src_folder download/02.Validation/ --name valid
python convert_data.py --src_folder download/03.Test/ --name dev

If you want to know statistics of Datasets, Use the script below

python show_data_statistics.py 

Training

If you want to change parameters, Edit the config file below

"ko": {
    "pretrained_model": "monologg/koelectra-base-v3-discriminator",
    "pkl_path": "pkl/koelectra-base-v3-discriminator",
    "pkl_name": "data.pkl",
    "table_path": "data/nia/tables.json",
    "train_path": "data/nia/train.json",
    "valid_path": "data/nia/valid.json",
    "dev_path": "data/nia/dev.json",
    "patience": 5,
    "learning_rate": 1e-6,
    "table_concat": false,
    "use_fixed_max_num": false,  
    "max_num": {
        "table_id" : 2,
        "table_num" : 2,
        "column_num" : 48,
        "select" : 6,
        "groupby" : 2,
        "orderby" : 2,
        "where" : 4,
        "having" : 1,
        "spc_id" : 2
    }
  },

Train with RYANSQL model

python main_ko.py --train --save_ckpt {SAVE_CKPT} --batch_size {BATCH_SIZE} --device {DEVICE}
ex) python main_ko.py --train --save_ckpt save_model/best_ckpt --batch_size 16 --device 0

Inference

Inference with RYANSQL model using evaluation script from Spider

./evaluate.py --save_ckpt {SAVE_CKPT} --batch_size {BATCH_SIZE} --device {DEVICE}
ex) ./evluate.py --save_ckpt save_model/best_ckpt --batch_size 16 --device 0

You can get csv file below

{SAVE_CKPT}/best_ckpt/predictions.csv

License

RYANSQL은 Apache-2.0 라이선스 하에 공개되어 있습니다. 모델 및 코드를 사용할 경우 라이선스 내용을 준수해주세요. 라이선스 전문은 LICENSE 파일에서 확인하실 수 있습니다.