Welcome to DATE, a LLM-based tabular data generator for heterogenous data.
Tabular data generation has become increasingly essential for enabling robust machine learning applications, which require large-scale, high-quality data. Existing solutions leverage generative models to learn original data distributions. However, real-world data are naturally heterogeneous with diverse distributions, making it challenging to obtain a universally good model for diverse data generation. To address this limitation, we introduce Diversity-Aware Tabular data gEnerator (DATE), a framework that (i) prepares high-quality and distributionally distinct examples for in-context learning by effectively partitioning the original heterogeneous data into multiple diverse subsets; (ii) harnesses Large Language Models (LLMs) to explore the diversity of the partitioned distribution with decision tree reasoning as feedback, generating high-quality labeled data for each subset.
We present our Source Code and Appendix for users: For detailed proofs and external experiments in the original paper, please read our appendix.
To obtain the public data used in our experiment, please refer to the IEEE Dataport. We also public it on Huggingface as Heterogenous_Synthesis_Benchmark
For running Module 1: DGR-based Prompt Designing, please refer to ModelShare_with_DSR_final.py.
For running Module 2: Distribution-Specific Generation, please refer to dataGeneration.py.
Files with the test_ prefix are test files. For example, (test_forward)[test_forward.py] is a testing file based on forward-greedy selection.