hitsz-ids/synthetic-data-generator

Leverages multiple synthesis approaches including GAN-based models (CTGAN, GaussianCopula), LLM-based generation for zero-shot synthesis, and statistical methods, with automatic column relationship detection to improve data quality. Features a modular Data Processor system for type conversion and preprocessing (datetime, null handling) with pluggable architecture, plus optimized memory efficiency for billion-scale datasets. Integrates differential privacy and anonymization capabilities alongside metadata inference for both single and multi-table scenarios.

2,409 stars.