Controllable Table Data Synthesis with Natural Language and Knowledge Base

This project provides a pipeline for generating structured table datasets using natural language descriptions and an optional knowledge base for controlled constraints. The workflow consists of three sequential steps:

1. generate_json.py - Accepts a natural language description of the desired table dataset and generates a JSON file representing column relationships.
2. generate_python.py - Takes the JSON file as input, optionally incorporating a knowledge base to enforce controlled constraints on the synthesized data.
3. generate_dataset.py - Uses the processed JSON and constraints to generate the final dataset.

Workflow

1. Define the dataset: Run generate_json.py with a textual description of the table data you want to synthesize.
2. Process constraints: Use generate_python.py to process the JSON output, optionally integrating domain-specific knowledge for enhanced control.
3. Generate the dataset: Execute generate_dataset.py to synthesize the structured data.

File Descriptions

generate_json.py

Functionality:

• Accepts a text description of the target dataset.
• Produces a JSON file defining column relationships.

Usage:

python generate_json.py --description "Your dataset description here"

generate_python.py

Functionality:

• Processes the JSON file generated in the first step.
• Optionally integrates a knowledge base to impose constraints on the synthesized data.

Usage:

python generate_python.py --json_file generated_structure.json [--knowledge_base knowledge_base.json]

generate_dataset.py

Functionality:

• Generates the final table dataset based on the structured JSON and optional constraints.

Usage:

python generate_dataset.py --json_file processed_structure.json

Dependencies

The project requires the following Python libraries. It is recommended to use Python 3.8 or later.

pip install -r requirements.txt

Ensure all dependencies are installed before running the scripts.

File Structure

Project Root
│   README.md
│   requirements.txt
│
└───scripts
    │   generate_json.py
    │   generate_python.py
    │   generate_dataset.py

Notes

• Ensure that all scripts are executed in the correct sequence for proper dataset synthesis.
• The knowledge base is optional but improves data generation control.
• Refer to inline comments within each script for additional details on parameters and configurations.

Contributors

For any issues or suggestions, feel free to reach out to the project maintainers.

---

Thank you for using this project! We hope it helps with your work.