Synthetic Data & Privacy
synthpriv
[ Open Source ][ Alpha ]
Tabular data synthesis preserving differential privacy
synthpriv generates synthetic tabular data under formal differential privacy guarantees: DP-SGD and pure-DP generators, marginal ECDFs with calibrated noise, and an epsilon-vs-utility sweep so you can see exactly what privacy costs you — with an HTML report for every run.
// Features
- Multiple generators: dp-gan (DP-SGD, AC-GAN conditioning) and dp-copula (pure DP Gaussian copula)
- DP marginal ECDFs with Laplace noise, with budget split between training and marginals
- Epsilon ↔ utility sweep — see the real privacy/utility tradeoff curve
- Benchmark against non-DP SDV baselines with utility gap reporting
- HTML privacy/utility report, model serialization, and DP integrity assertions out of the box
// Technologies
PythonSDVOpacus