Statistical forgery: generating fake data that actually behaves like the real thing

What it does

Copulas is a Python library for modeling multivariate distributions and sampling synthetic data from them. Feed it a table of numerical data, pick a copula function (Gaussian, Archimedean, or Vine), and it learns the joint statistical behavior — then generates new rows that mimic the original relationships, not just column-by-column histograms.

The interesting bit

The library exposes the full internals of its models, so you can inspect and tweak learned parameters rather than treating it as a black box. It also ships with built-in 1D, 2D, and 3D visualizations to eyeball whether your synthetic data actually matches the real stuff — a sanity check that too many generative tools skip.

Key highlights

• Supports Gaussian, Archimedean, and Vine copulas for different dependency structures
• Built-in visualization suite: histograms, 2D scatterplots, and 3D scatterplots for comparing real vs. synthetic
• Full parameter access: inspect and manually adjust fitted model internals
• Part of the broader SDV (Synthetic Data Vault) ecosystem from DataCebo, originally started at MIT’s Data to AI Lab in 2018
• Available via pip and conda-forge

Caveats

• Marked as “Pre-Alpha” development status, so expect API instability
• README focuses on numerical data; unclear how it handles categorical columns or mixed types without preprocessing

Verdict

Worth a look if you need statistically grounded synthetic tabular data and want visibility into how the sausage is made. Skip it if you need production-grade stability today or deep learning-based generation out of the box — the SDV umbrella package may be the better entry point.