synthpriv

Tabular data synthesis preserving differential privacy: generators (dp-gan with DP-SGD, dp-copula with pure DP), DP marginal ECDFs, benchmark and budget sweep, and an HTML privacy/utility report. Built on SDV, SDMetrics and Opacus.

Note

Synthetic data without DP is not guaranteed anonymization. Formal protection only comes from training with a DP mechanism (DPSGD) and verifying it with assert_dp().

Install with:

pip install synthpriv

User guide

• Quick start
  • Installation
  • Python API
  • CLI
  • Reproducible demo
• Differential privacy
  • Mechanics
  • Mechanisms
  • Generators
  • dp-copula
  • dp-gan
  • Budget split
  • Limitations
  • Privacy note
• Sweep and benchmark
  • Metrics
  • Epsilon vs utility sweep
  • Benchmark: DP vs non-DP baselines
• Command-line interface
  • synthpriv generate
  • synthpriv sample
  • synthpriv evaluate
  • synthpriv sweep
  • synthpriv benchmark
  • synthpriv dpcheck

Reference

• API reference
  • Pipeline
  • Sweep
  • Benchmark
  • Report
  • Core
  • Generators
  • Differential privacy
  • DP generators
  • Metrics
  • Utilities
  • Command-line interface
• Changelog
  • Unreleased
  • 0.2.1 - 2026-09-21
  • 0.2.0 - 2026-09-15
  • 0.2.0b1 - 2026-09-15
  • 0.1.0 - 2026-09-08

Indices

• Index

• Module Index

• Search Page

Site analytics

This documentation site uses Google Analytics 4 to count page views, with IP anonymization enabled. No personal data is collected and no synthpriv cookie is set; the measurement ID is a repository setting (GOOGLE_ANALYTICS_ID) and can be turned off with GOOGLE_ANALYTICS_ID=off.