Welcome to the synthyverse! An extensive ecosystem for synthetic data generation and evaluation in Python. Read the docs for in-depth usage. The synthyverse is a work in progress. Please provide any suggestions through a GitHub Issue.

Features

• Tabular synthetic data generators. Use low-level generators directly, or wrap them with shared preprocessing through SynthyverseGenerator.
• Evaluation metrics. Compare synthetic data with fidelity, utility, and privacy metrics through individual metric classes or TabularMetricEvaluator.
• Benchmarking workflows. Train, sample, evaluate, and save benchmark artifacts with TabularSynthesisBenchmark.
• Shared preprocessing. Reuse DataProcessor for missing-value handling, schema restoration, and column constraints.

Installation

Install synthyverse from PyPI:

pip install synthyverse

The base package is MIT licensed and does not install the ctgan package. CTGANGenerator and TVAEGenerator are only available when installing the optional CTGAN extra:

pip install "synthyverse[ctgan]"

That extra installs the ctgan dependency, which is distributed under the Business Source License. Review that license before using CTGAN or TVAE functionality.

Third-party attribution, license, NOTICE, and modification details are listed in THIRD_PARTY_NOTICES.md.

For local development from a clone:

pip install -e .

Usage

Use a high-level wrapper when you want preprocessing and schema restoration handled for you:

from synthyverse.generators import SynthyverseGenerator

generator = SynthyverseGenerator(
    "univariate",
    missing_imputation_method="median",
    random_state=42,
)
generator.fit(X, discrete_features=["category", "target"])
X_syn = generator.generate(1000)

Use the lower-level APIs when you want explicit control over preprocessing, generator fitting, metrics, or benchmarking. See the docs for complete examples.