# SDV (Synthetic Data Vault)

> Python library for generating and evaluating synthetic tabular, multi-table, and sequential data using machine learning models.

The Synthetic Data Vault (SDV) is a Python library from DataCebo for creating tabular synthetic data. It learns patterns from real data with machine learning models and generates new data that emulates them, for single tables, multiple connected tables, or sequential data. The project began at MIT's Data to AI Lab in 2016, and DataCebo was created in 2020 to grow it.

## What It Is

SDV is a library installed with pip or conda. Users load real data along with metadata that describes column types and keys. They then fit a synthesizer to that data and sample as many synthetic rows as they need. The documentation says it is designed to run on-prem with standard CPUs.

## Modeling, Sampling and Evaluation

SDV offers several models, ranging from classical statistical methods such as GaussianCopula to deep learning methods such as CTGAN. The documentation organizes synthesizers into single table, multi table, and sequential types. Sampling covers realistic data generation, conditional sampling, and export.

In the README example, sensitive columns are anonymized, statistical patterns in the other columns are preserved, and primary and foreign key relationships stay intact. Evaluation tools produce diagnostic and quality reports with column shape and column pair trend scores. They also provide visualizations that compare real and synthetic data, and the docs include a privacy evaluation section.

## Customization

Users can adjust data preprocessing, choose anonymization options, and define business rules as logical constraints. The docs also describe Constraint-Augmented Generation and integration with databases and local data.

## Editions and Licensing

SDV Community is the public library, released under the Business Source License 1.1. That license is not an OSI open source license, and it restricts use in a commercial Synthetic Data Service. SDV Enterprise is available to licensed users. The docs state it adds scalable synthesizers for large numbers of interconnected tables, more advanced preprocessing and enhanced algorithms, and easier data source integration and deployment. The docs also list an SDV Agent section.

## Features
- Single table, multi table, and sequential synthesizers
- Statistical and deep learning models (GaussianCopula, CTGAN)
- Quality and diagnostic reports
- Real vs synthetic data visualization
- Privacy evaluation
- Constraints and business rules
- Data preprocessing control
- Anonymization of sensitive columns
- Conditional sampling
- Database and local data integration
- Runs on-prem on standard CPUs

## Platforms
API, DEVELOPER_SDK, CLI

## Pricing
Free

## Version
v1.38.5

## Links
- Website: https://docs.sdv.dev/sdv
- Documentation: https://docs.sdv.dev/sdv
- Repository: https://github.com/sdv-dev/SDV
- EveryDev.ai: https://www.everydev.ai/tools/sdv