Sdv
Synthetic data generation for tabular data
Last verified:
What is Sdv?
The Synthetic Data Vault (SDV) is a Python library designed to be your one-stop shop for creating tabular synthetic data. It uses machine learning to analyze real data and create fully synthetic datasets that mimic the original format and mathematical properties. SDV supports single table, sequential (time series), and multi-table (relational) data modalities, making it versatile for various data types.
Key features include the ability to train your own generative AI model with 14+ AI algorithms, evaluate and visualize synthetic data quality using SDMetrics library, and customize synthesizers with constraints, data preprocessing options, and anonymization features. The platform offers constraint-augmented generation that ensures synthetic data meets business rules 100% of the time, with predefined constraints like Inequality, Range, FixedCombinations, and the ability to program custom constraints.
SDV is designed for data scientists, machine learning engineers, software developers, and enterprises that need synthetic data for testing applications, augmenting datasets, training ML models, removing bias, or sharing data safely while maintaining privacy. The library runs on-prem with standard CPUs and is owned and maintained by DataCebo, continuing the project originally created at MIT's Data to AI Lab in 2016.
The platform includes evaluation tools like Diagnostic Reports for data validity checks and Quality Reports for statistical similarity, plus visualization capabilities with 1D/2D bars, scatterplots, and heatmaps. Users can also implement differential privacy guarantees with specialized synthesizers like DPGC and DPGCFlex.
Sdv pricing
Pricing model: Freemium
SDV offers three plans: (1) SDV Community is free/no cost with 3-5 tables, 4 data types, 6 basic models, 5 constraint strategies, CSV/dataframe support, quality metrics, public benchmarks, and support via GitHub/Slack. It uses Business Source License with limited commercial use. (2) SDV Enterprise uses tiered pricing (contact sales) with unlimited tables, 30+ data types, 14+ models, 20+ constraint strategies, ROI metrics, private leaderboards, dedicated support via discourse, and optional SDV Bundles add-ons. (3) SDV Custom offers custom pricing for enterprise scale with unlimited tables/data types, use case specific models, on-demand database connector, all accelerators, customized evaluation frameworks, and 24/7 Customer Success & Professional Services. Credit card payments accepted (Visa, MasterCard, AMEX, Discover). No refunds after purchase. Special discounts available for nonprofits and educational institutions.
Sdv pros
- Free SDV Community tier available for getting started
- On-prem deployment keeps data within your environment
- Works on standard CPUs without GPU requirements
- Supports single-table, multi-table, and sequential data
- 14+ AI models including GaussianCopula, HMA, PAR, XGC
- 20+ constraint handling strategies for business rules
- 30+ data types supported in Enterprise plan
- Auto-detect metadata from CSVs, DataFrames, and databases
- Built-in evaluation with SDMetrics library included
- Differential privacy guarantees available with DPGC synthesizers
- Custom constraint programming for single and multi-table
- Reversible data transformers for anonymization
- Conditional sampling for fixed values in synthesis
- Direct database connectors for importing/exporting data
- Active community support via GitHub and Slack
- Diagnostic and Quality Reports for data validation
- Plotly visualizations for comparing real vs synthetic data
Sdv cons
- SDV Community limited to 3-5 tables only
- Community plan supports only 4 data types
- Community has only 6 basic models available
- Only 5 constraint strategies in free tier
- Business Source License restricts some commercial use
- Enterprise pricing requires contacting sales (no public pricing)
- No refunds once Enterprise purchase is made
- Downgrading SDV versions is not supported
- Older synthesizers (pre-v1.10.0) may not allow upgrades
- Plotly charts may not render inline in VSCode/JupyterLab
- Reject sampling can slow down the sampling process
- ConstraintsNotMetError if real data violates constraint rules
- Limited synthetic data accelerators in Community tier
- No database connector in Community plan
- Public support only (no dedicated support) for Community
Frequently asked questions about Sdv
What is SDV and what does it do?
The Synthetic Data Vault (SDV) is a Python library designed to be your one-stop shop for creating tabular synthetic data. It uses machine learning to analyze your real data and creates fully synthetic datasets that mimic the original format and mathematical properties. SDV supports single table, sequential (time series), and multi-table (relational) data, allowing you to generate any amount of synthetic data on-demand for testing, ML training, data augmentation, or secure data sharing.
Is SDV free to use?
Yes, SDV Community is free at no cost. It allows you to get started with synthetic data using 3-5 tables, 4 data types, 6 basic models, and 5 constraint handling strategies. SDV Community is distributed under the Business Source License with limited commercial use and is designed as a platform to evaluate SDV for a use case before deciding to upgrade to a paid Enterprise plan.
How do I install SDV?
SDV Community is installed using pip: pip install sdv. Alternatively, you can use conda. SDV is developed on Python 3.9-3.13 and we recommend using a virtual environment (such as virtualenv) to avoid conflicts. After installation, verify it worked by importing sdv and printing sdv.version.community. For SDV Enterprise, you use credentials with sdv-installer: pip install sdv-installer --upgrade, then run sdv-installer install --upgrade with your username and license key.
What AI models does SDV offer?
SDV offers 14+ AI models in Enterprise (6 basic models in Community). Models include GaussianCopula (statistical AI), XGC (advanced Copula modeling), PAR (for sequential data), HMA (multi-table for limited tables under 5), HSA (multi-table for unlimited tables), Independent (multi-table for unlimited tables), SegmentSynthesizer (for segmented data), BootstrapSynthesizer (for data with few rows), and DPGC/DPGCFlex (for differential privacy guarantees). You can choose algorithms designed for single table, sequential, or multi-table data.
How do constraints work in SDV?
Constraints let you input business rules into your synthesizer ensuring synthetic data meets the constraint 100% of the time. SDV uses three strategies: Transformation (transforms data so models learn the constraint, then reverses it), Reject Sampling (throws away rows violating constraints, may slow sampling), and Algorithmic Injection (complex CAG patterns with their own algorithms). Predefined constraints include Inequality, Range, FixedCombinations, FixedIncrements, OneHotEncoding, ChainedInequality, and you can program custom constraints in Python.
Can I use SDV with my database?
SDV Community supports CSVs and DataFrames only. Direct database connectors for importing real data and exporting synthetic data are available in SDV Enterprise and SDV Custom plans. Enterprise includes on-demand database connector as part of integrations, while Custom plan includes it as a standard feature. Auto-detect metadata based on your database is also an Enterprise feature.
How do I evaluate synthetic data quality?
SDV includes the SDMetrics library for evaluation. You can generate a Diagnostic Report for basic data validity checks (single and multi-table) and a Quality Report for statistical similarity between real and synthetic data (single and multi-table). Enterprise adds ROI metrics and frameworks, private leaderboards, and dataset integration. You can also verify differential privacy, measure DisclosureProtection, and use visualizations including 1D/2D bars, scatterplots, and heatmaps to compare data.
What is the difference between SDV Community and Enterprise?
SDV Community is free with 3-5 tables, 4 data types, 6 basic models, 5 constraint strategies, CSV/dataframe support only, quality metrics, public benchmarks, and public support (GitHub/Slack). SDV Enterprise has tiered pricing with unlimited tables, 30+ data types, 14+ models, 20+ constraint strategies, database connectors, all accelerators, ROI metrics, private leaderboards, dedicated discourse support, and optional SDV Bundles add-ons. Enterprise also has advanced preprocessing, deeper data understanding, and enhanced AI algorithms.
Does SDV support differential privacy?
Yes, SDV offers differential privacy guarantees through specialized synthesizers: DPGC and DPGCFlex Synthesizers create synthetic data with differential privacy guarantees. SDV also includes DP transformers like DPLaplaceNoiser, DPTimestampLaplaceNoiser, DPResponseRandomizer, DPWeightedResponseRandomizer for adding noise to guarantee differential privacy, and DPECDFNormalizer/DPDiscreteECDFNormalizer for normalizing columns while guaranteeing differential privacy. You can verify the differential privacy of any synthesizer algorithm.
What support is available for SDV users?
SDV Community users get support via public channels - GitHub and Slack. SDV Enterprise users get dedicated support via discourse with priority support. SDV Custom users receive Dedicated Customer Success & Professional Services with 24/7 customer support. For technical assistance with installation, troubleshooting, or general questions, all users can use the DataCebo Forum. If you need help with licensing or pricing, contact [email protected] or reach out through the contact us page. Special discounts are available for nonprofits and educational institutions.