PyMC's Big Adventure: Lessons Learned from the Development of Open-source Software for Probabilistic Programming

Christopher J. Fonnesbeck

Keywords: pymc, python, open source, sustainability, collaboration, probabilistic programming

TL;DR: I will share some of the lessons learned since 2003, and provide a summary of best practices and potential pitfalls associated with growing and sustaining a community-driven scientific software project.

Abstract:

The PyMC project is a team of open source developers devoted to the development of software for
applied Bayesian statistics and probabilistic machine learning. Broadly, our objective is to produce
Python implementations of state-of-the-art methods that can be used by a wide range of non-expert
analysts, thereby democratizing probabilistic programming and putting powerful Bayesian methods
in the hands of those who need them most: economists, astronomers, epidemiologists, ecologists,
and more. Our current product, PyMC3, allows users to implement arbitrary probabilistic models
using a high-level API that is analogous to specifying a model on a whiteboard.
The first version of PyMC was created over fifteen years ago, as a solo academic project. Its original
purpose was to provide an open-source alternative to WinBUGS, a proprietary single-platform
software package that dominated applied Bayesian analysis at the time. PyMC2 (Patil et al. 2010),
released in 2009, represented a transition to a high-performance code base, using optimized Fortran
extensions, and was the first truly collaborative release, based largely on the efforts of a team of three
developers. The current major release, PyMC3 (Salvatier et al. 2010), is the product of a complete
re-write of the code base, moving to Theano (Bergstra et al. 2010), a deep learning framework, as
the computational backend. This allowed for the next generation of Bayesian model fitting methods
to be implemented, including gradient-based (Hamiltonian) Monte Carlo and variational inference.
Today, PyMC3 is supported by over a dozen core developers, and is used widely to support academic
and industry research and decision-making.
Most of PyMC’s history was characterized by substantial contributions by a few talented but
ephemeral developers. The relative success and stability that the project enjoys today is the result
of the hard work of a small but enthusiastic core of users and contributors, the emergence of tools
for building software communities online, namely GitHub, StackOverflow, and Slack, and a small
dose of good fortune. I will share some of the lessons learned since 2003, and provide a summary
of best practices and potential pitfalls associated with growing and sustaining a community-driven
scientific software project.

Decision: accept