SDV (Synthetic Data Vault)

// Overview

SDV (Synthetic Data Vault) is a Python library for creating tabular synthetic data. It uses machine learning models to learn the patterns in your real data, then generates new tables that follow those same statistical patterns without copying the original rows.

It is meant for data scientists and engineers who need realistic test data, want to share datasets without exposing sensitive values, or need extra training data. SDV handles single tables, multiple connected tables, and sequential (time-based) data.

Within the synthetic data category, SDV covers the full workflow: it describes your data with metadata, fits a synthesizer, samples new rows, and then evaluates how close the synthetic data is to the real data with a quality report.

// What it does

• Multiple models, from the statistical GaussianCopula to the deep-learning CTGAN
• Generates data for single tables, multiple connected tables, or sequential tables
• Quality reports and column plots to compare synthetic data against the real data
• Built-in anonymization for sensitive columns such as emails, addresses, and credit card numbers
• Preserves keys and relationships, including unique primary keys and primary/foreign key links
• Logical constraints to encode business rules during data processing

// Getting started

Install SDV with pip, load a demo dataset, fit a synthesizer, and sample synthetic rows.

pip install sdv

from sdv.datasets.demo import download_demo

real_data, metadata = download_demo(modality='single_table', dataset_name='fake_hotel_guests')

from sdv.single_table import GaussianCopulaSynthesizer

synthesizer = GaussianCopulaSynthesizer(metadata)
synthesizer.fit(data=real_data)

synthetic_data = synthesizer.sample(num_rows=500)

from sdv.evaluation.single_table import evaluate_quality

quality_report = evaluate_quality(real_data, synthetic_data, metadata)

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Generate safe test and demo data that mirrors production without exposing real customer values
• Share or publish a dataset by replacing sensitive columns with anonymized synthetic ones
• Create extra training data when a real dataset is small or imbalanced
• Reproduce multi-table or time-series data while keeping keys and relationships intact

// How SDV (Synthetic Data Vault) compares

SDV (Synthetic Data Vault) alongside other open-source synthetic data tools AI/TLDR tracks, ranked by GitHub stars.