SDV (Synthetic Data Vault)

The Synthetic Data Vault started at MIT's Data to AI Lab in 2016 and remains the most established Python toolkit for tabular synthetic data: it learns generative models of single tables, multi-table relational schemas with foreign keys intact, and sequential data, then samples new rows that preserve the statistical shape of the original. Synthesizers range from the fast statistical GaussianCopula to deep learning approaches such as CTGAN, and built-in evaluation scores quality on column shapes and column pair trends, with plots comparing real against synthetic distributions. Typical uses are sharing realistic data where the real thing is sensitive, generating test data for software, and augmenting small datasets. It installs with pip or conda and runs locally on CPU, with GPU optional for the neural synthesizers. DataCebo, the company formed around the project in 2020, maintains it under the Business Source License, which allows free use but is not a traditional open source license and reserves certain commercial rights, with paid enterprise offerings on top. The repository holds about 3,500 GitHub stars.