LCATS

LCATS is an LLM-guided framework for constraint-aware tabular data synthesis. It combines executable constraint discovery, diffusion-based generation, diagnostic bucketing, and adaptive weighted repair to produce synthetic tables that are both statistically faithful and logically valid.

Installing Dependencies

Python version: 3.10

Create the TabSyn environment

conda create -n tabsyn python=3.10
conda activate tabsyn
pip install -r requirements-tabsyn.txt

Create the SynthCity environment

conda create -n synthcity python=3.10
conda activate synthcity
pip install -r requirements-eval.txt

API Configuration

export OPENROUTER_API_KEY="your-key"

Demo

Run LCATS from the repository root:

python LCATS/run_lcats.py --dataname insurance