genpark-wukong-data-synthesizer

GenPark Wukong Data Synthesizer is a zero-dependency differential privacy and synthetic data validation engine implementing the Laplace mechanism for privacy budget $\epsilon$, 1D Wasserstein Earth Mover's Distance distribution fidelity scoring, and Frobenius norm correlation preservation matrices.

flowchart LR
    A[Real Numeric Column] --> B[Laplace Mechanism Perturber]
    B --> C[Epsilon-Differentially Private Dataset]
    D[Real vs Synthetic Quantiles] --> E[Wasserstein W1 Distance Solver]
    E --> F[Empirical Distribution Fidelity Score]
    G[Pairwise Correlation Matrices] --> H[Frobenius Norm & MAE Comparator]
    H --> I[Structural Preservation Score]

Features

• Laplace Differential Privacy: Synthesizes noise scaled to $b = \Delta f / \epsilon$ with rigorous 95% confidence intervals.
• Wasserstein Distance Metric: Measures exact Earth Mover's Distance $W_1$ normalized to empirical interquartile ranges.
• Correlation Matrix Fidelity: Verifies that synthetic feature covariance matrices preserve underlying multivariate relationships.

Quickstart

from client import GenParkWukongDataSynthesizer

synth = GenParkWukongDataSynthesizer()
res = synth.generate_laplace_differential_privacy_noise(100.0, epsilon=1.0)
print("Privatized Value:", res["privatized_value"])

License

MIT