cmdstanpy

CmdStanPy is a Python interface to CmdStan, the command-line interface for Stan, a platform for Bayesian statistical modeling. It provides a Pythonic API to compile Stan models, supply data, run inference via CmdStan (sampling, optimization, or variational inference), and extract results without leaving Python.

The package operates as a thin wrapper around the CmdStan executable. When using CmdStanPy, a Stan model file is loaded to create a CmdStanModel object; the model can be compiled and then run with data and sampling parameters. The resulting outputs are written to a directory as text and CSV files, which CmdStanPy can parse into Python data structures for analysis.

Key features include: model compilation and management, data handling, support for sampling with multiple chains and warmup periods, interfaces for optimization, and utilities to read posterior samples and summary diagnostics into pandas dataframes and, with additional packages, in ArviZ-compatible formats. It supports cross-platform operation on Windows, macOS, and Linux, and relies on a CmdStan installation on the host system.

Installation typically requires pip install cmdstanpy and a CmdStan installation on the machine; the path to CmdStan can be configured via environment variables or model-specific parameters. Usage examples involve creating a CmdStanModel from a Stan file, calling the sample method with data and sampling options, and accessing the fit object to obtain draws and diagnostics.

CmdStanPy is open-source and maintained as part of the Stan ecosystem, designed to fit into Python workflows for reproducible Bayesian inference.