Every synthetic data project starts with the same question: Which synthesizer should I use?
The answer: it depends. The right choice depends on your data, schema, and constraints.
SDV Enterprise offers 13 synthesizers across single-table, multi-table, and sequential data. Some are general-purpose, while others are built for specific problems. They span a range of approaches—from classical statistical methods to GANs and variational autoencoders—each designed for different scenarios.
To make selection easier, we created a cheatsheet that maps each synthesizer: the type of data it supports, the algorithm behind it, what it is designed for, and where it is available.
A few patterns worth knowing:
Single-table models
→ GaussianCopula: the recommended starting point — fast, customizable, and a strong baseline
→ Neural network synthesizers (CTGAN, TVAE): trade speed for flexibility when modeling complex distributions
→ XGC: extends the copula approach with more flexible distributions and faster runtime
Multi-table synthesizers
→ HMA: exhaustive modeling of parent–child relationships across the schema
→ Independent: optimized for speed and scale
→ HSA: balances speed with preserving essential cross-table relationships
Specialty synthesizers
→ Bootstrap: designed for small datasets
→ Segment: built for highly segmented data
→ DayZ: designed for day-zero data generation
→ DPGC: enables differentially private synthetic data generation
→ PAR: models multi-entity sequential data
Save this cheatsheet. Before your next project, take a look and see if it helps you make that first decision.
SDV Enterprise cheat sheet