PyMC

Bayesian estimation, particularly using Markov chain Monte Carlo (MCMC), is an increasingly relevant approach to statistical estimation. However, few statistical software packages implement MCMC samplers, and they are non-trivial to code by hand. PyMC is a python module that implements a suite of MCMC algorithms as python classes, and is extremely flexible and applicable to a large suite of problems. PyMC includes methods for summarizing output, plotting, goodness-of-fit and convergence diagnostics.

Download

Source code and binary packages for the 2.0 release are now available at the Python Package Index. If a binary package is not available for your platform, building PyMC should pose no particular problem, requiring only a C and Fortran compiler. Build instructions can be found in the user guide included in the source distribution.

Access to the source code is available at GitHub (the SVN repository on Google Code has been deprecated). We recommend users build PyMC from the development code rather than using the released version, as it contains bug fixes and features not available in the current release, and tends to be pretty stable and reliable.

Macintosh (OSX 10.6) users may obtain a current build of PyMC as part of the Scipy Superpack, which includes all dependencies in an easy-to-install script.

Previous users of the 1.3 version should note that the syntax used to define statistical models has changed significantly. While it is always difficult to break backward-compatibility, the changes bring drastic improvements in performance and flexibility.

Examples

To get an idea of what a PyMC2 model looks like, we have provided a few examples:

• Simple occupancy model using latent states
• Waterfowl band recovery model
• Simple pricing model
• Hierarchical Poisson failure rates
• Fisheries surplus production model (updated Aug. 2009)
• Salamander occupancy estimation model
• Simple probit regression model
• Exponential model for melanoma survival data
• Hierarchical disease dynamics model (from Zipkin et al. 2010)

For users familiar with BUGS, here are a few examples that are translated directly from BUGS models; the original code is included in each file as a docstring:

• Koala sighting model (from Link & Barker 2009)
• Conditional multinomial mark-recapture model (from Link & Barker 2009)
• Unconditional multinomial mark-recapture model (apparently not possible in BUGS)
• Simple example of Bayes factor calculation

User-contributed tutorials and recipes can be found in in the wiki. More examples can be found in the examples folder in the source tree.

Also, the user guide contains a tutorial section.

Support

The PyMC User’s Guide contains detailed installation instructions, as well as some MCMC theoretical background and a tutorial on using PyMC. The user's guide is available in the downloads section.

For help, questions or suggestions on a particular topic related to PyMC, please visit the PyMC Google Group.

If you wish to file a bug report or suggest enhancements to PyMC, please submit an issue at our issues page (which replaces our old issues tracker.

Citation

When referencing PyMC in your publication, the appropriate citation is:

Patil, A., D. Huard and C.J. Fonnesbeck. 2010. PyMC: Bayesian Stochastic Modelling in Python. Journal of Statistical Software, 35(4), pp. 1-81.