Practicing Trustworthy Machine Learning

Appendix A. Synthetic Data Generation Tools

For domain-specific data generation tools, you have a variety of options.

In the spirit of GANs and flow-based models, there are plenty of projects that train generative models on real-world data and then use the generators as the source of synthetic data. Table A-1 lists several GAN-based methods.

Table A-1. Data-driven methods and tools^a

Methods and tools	Description	Further reading	Type
CTGAN	A GAN-based data synthesizer that can generate synthetic tabular data with high fidelity	“Modeling Tabular Data Using Conditional GAN”	Tabular
TGAN	Outdated and superseded by CTGAN	Tabular
gretel	Creates fake synthetic datasets with enhanced privacy guarantees	Tabular
WGAN-GP	Recommended for training the GAN; suffers less from mode-collapse and has a more meaningful loss than other GAN-based data generation tools	“On the Generation and Evaluation of Synthetic Tabular Data Using GANs”	Tabular
DataSynthesizer	“Generates synthetic data that simulates a given dataset and applies DP techniques to achieve a strong privacy guarantee”	Tabular
MedGAN	“[A] generative adversarial network for generating multilabel discrete patient records [that] can generate both binary and count variables (i.e., medical codes such as diagnosis codes, medication codes, or procedure codes)”	“Generating Multi-label Discrete Patient Records Using Generative Adversarial Networks”	Tabular
MC-MedGAN (multi-categorical GANs)	Produces synthetic data instances ...