Overview¶
CmdStanPy is a lightweight interface to Stan for Python users which provides the necessary objects and functions to do Bayesian inference given a probability model and data. It wraps the CmdStan command line interface in a small set of Python classes which provide methods to do analysis and manage the resulting set of model, data, and posterior estimates. It is lightweight in that it uses minimal memory beyond the memory used by CmdStan. CmdStanPy runs CmdStan, but only instantiates the resulting inference objects in memory upon request. Thus CmdStanPy has the potential to fit more complex models to larger datasets than might be possible in PyStan or RStan.
CmdStan is a file-based interface. CmdStanPy manages the Stan program files and the CmdStan output files. The user can specify the output directory for the CmdStan outputs, otherwise the files will be written to a temporary filesystem which persists throughout the session. This allows the user to test and develop models prospectively, following the Bayesian workflow.