We propose investigating the formation of a Statistical Python OSE that will catalyze the development of a robust set of statistical software for Python, and build a vibrant ecosystem of statisticians, domain practitioners, and software developers around it. Such an OSE will provide a central coordinating mechanism, through which knowledge around software engineering and community best practices can be aggregated and disseminated to statistics method developers and the researchers that apply those methods. To inform the design of the OSE, several meetings will be held that bring together statisticians, researchers, developers, students, and teachers. We will investigate existing OSEs, both in the Python and R worlds, and learn from each which effective social and technological practices to adopt. To anchor the effort, two pilot projects, with different scopes, target audiences, and levels of maturity will be audited to determine how they should be modified to comply with modern software engineering and community governance best practices. Through this audit, we will discover the nature of activities that are valuable to support through the OSE, and that are beneficial to the statistics developer community. Our vision for the OSE is that it will catalyze growth of a vibrant, inclusive statistical developer community in Python, that will also have repercussion more broadly as it builds bridges between scientific Python and other computational communities.
This proposal addresses two gaps in the statistical software ecosystem. On the one hand,
there are many small groups or individual developers of statistical packages who lack the training
in best (or better) practices in software engineering needed for their open-source projects to survive. On the other hand,
the scientific Python community, which adheres closely to such standards, resulting in larger, well-engineered software
packages, lacks participation of statistical methods developers. This proposal seeks to
learn, and document how, to bridge this gap, starting with YAGLM and ISLP as pilot
projects. From this knowledge, the OSE will be able to derive and propagate
models of best technical and social practices, resulting in better engineered software in the
statistical community, improved software development skills among
statisticians, and wider participation of statisticians in scientific Python.
The proposed OSE will act to establish
better software engineering practices in the statistical community,
provide onboarding pathways for involving young researchers, while documenting and implementing
healthy and inclusive community practices that can be replicated in other communities.
This is expected to result in more statistical libraries being available to researchers using Python,
as well as a cross pollination of best practices between R and Python.
\vfill
Keywords: MPS; Statistics; Generalized Linear Models; Statistical Learning; Science Education