PyStan

Pystan is a Python interface to Stan, a probabilistic programming language used for Bayesian inference. It enables Python users to define statistical models in Stan’s modeling language and to perform full Bayesian analysis using Stan’s sampling and optimization algorithms, such as Hamiltonian Monte Carlo and the No-U-Turn Sampler.

The library exposes a Pythonic API for working with Stan models. A user writes a model as a string or file, compiles it to a StanModel, provides a data dictionary that matches the model’s blocks, and then runs sampling or optimization. The output comprises posterior draws for each parameter, along with summaries and diagnostics such as R-hat and effective sample size.

Pystan has seen two major generations. PyStan 2 was the legacy interface that bundled the Stan C++ library, while PyStan 3 is a newer rewrite with a revised API and packaging. Users may encounter recommendations to consider alternative workflows (for example, CmdStanPy) in order to pair Stan models with different backends or newer Stan releases. In practice, PyStan remains one of the common ways to run Stan models directly from Python.

Typical usage involves defining a model, compiling it, preparing a data dictionary, performing sampling, and then extracting posterior samples for analysis in Python (e.g., with NumPy or pandas). The framework supports posterior predictive checks and model diagnostics, making it suitable for a range of Bayesian modeling tasks across disciplines.

Pystan is open-source and maintained by the Stan community. It is distributed via PyPI and interoperates with standard Python scientific tooling, with documentation available on the Stan project site and its GitHub repositories.