Bootstrapping a SciPy-related community project

lighting_talk.rst

Bootstrapping a SciPy-related community project

This document is a lightning talk: it only gives pointers, you need to Google and read references

Goal

Creating a community-driven project in Scientific Python. For the success of the project, garnering good developers is critical.

Preliminaries

• Join an existing project rather than creating one, if one exists in the scope that you are interested in.
• Discuss your project on the scipy-dev mailing list

Ingredients for success

• Success with developers == Success with users
• Bootstrap with the right people and grow
• Documentation is key for end-users and developers
• Tests are necessary
• Well defined goal and scope
• Keep bounds on technical difficulty
• Short release cycles
• Focus on not owning the project
• Communicate a lot (blog, twitter, mailing lists), but with a good signal-to-noise ratio

Technical recipe

• Use git + github:

  • Create an organization on github http://github.org
  • Learn git (large investment, you'll struggle).
  • Host the website on github

• Create a mailing-list (host it on http://librelist.com or google groups)
• Use Sphinx (http://sphinx.pocoo.org/) for the documentation and website
• Do many tests with nosetests
• Do a lot of code review using github pull request mechanisms. Try to have people reviewing your own code. Value criticism. Give positive and polite criticism.
• make a setup.py, package your code (see distutils documentation, http://packages.python.org/an_example_pypi_project/index.html and example packages listed below).
• Separate the code from the examples. Ideally generate documentation from examples (as in scikit-learn/scikit-image)
• start with pure Python code (after a year and some community building, you can introduce Cython).

Examples project

Inspect the social dynamics and technical choices of successful projects (list below reflects personal bias)

• Joblib (http://packages.python.org/joblib/) Very simple example of a project with tests, packaging and documentation
• scikit-learn (http://scikit-learn.org) and scikit-image (http://scikits-image.org/): mature projects with a striving social dynamic. In particular, read the developer guidelines for the scikit-learn: http://scikit-learn.org/dev/developers/index.html

Thanks for putting this together, you might want to add:

http://packages.python.org/an_example_pypi_project/index.html

which is an excellent starting point when starting a project!