Announcement of the grant being awarded

We are happy and proud to announce that the NASA ROSES 2020 program, specifically the Support for Open Source Tools, Frameworks and Libraries component, has accepted a proposal from the scientific Python community. The selection is for a 3-year, $385,385 per year grant, which will be split between the participating projects: scikit-learn, pandas, SciPy and NumPy.

Backstory

Once the NASA ROSES funding proposal opened, the working group began creating a proposal that answered the conditions of the program. The proposal was based on an earlier rejected NSF proposal. Many of the authors of the NSF proposal were also involved in the successful NASA ROSES proposal. The process of pulling the proposal together: laying out the scope of the program, splitting the grant writing into sections, and overall cooperation between the groups was in and of itself helpful to get many of the people at the core of scientific Python together.

Deliverables

This Gantt chart summarizes the primary deliverables of the project:

• Ongoing work with issue triaging & code review, maintenance, CI & packaging improvements
• Creating a joint infrastructure for running benchmarks and create both micro-benchmarks and more general benchmarks for the projects
• Use NumPy's new dtype infrastructure to create a flexible Unicode string type, and integrate that into pandas
• More NumPy SIMD performance improvements
• Move forward with Array API Standard adoption
• Extend the use of Numba UDFs in pandas
• Optimize pandas' memory usage, and implement Cython performance optimizations in sckit-learn
• Add support for CuPy and Dask to SciPy and scikit-learn via the Array API standard
• Add a framework for parallelization to SciPy
• Add large-scale optimization routines to SciPy

The goals are quite ambitious given the amount of funding. They leverage synergy between the projects and underlying standards to move the entire scientific Python community forward.

Who will be doing the work

The work will be executed by a mix of experienced maintainers and new talent recruited specifically for the project. Due to restrictions from the funders, research (which includes most or all work on new features) will be primarily executed by people in the United States, while the international team will take on the engineering and contributor tasks. Some of the work will be subcontracted to LANL and Cal Poly.

The grant will be administered via a committee. Dharhas Pothina, Quansight's CTO, is the PI. Leaders of the participating projects will advise and monitor activities and changes in scope to make sure they are aligned with what the projects need most.

Final thoughts

This is a significant milestone in the growing stream of funding flowing towards community-driven Open Source - and to scientific Python projects in particular. The grant is one of the first to specifically fund cross-project collaboration, in a way that will leverage common interests toward improving the entire ecosystem. Institutional funders are realizing that we are indeed a web of interlinked projects supported by a community of contributors, and we are thankful to NASA for the opportunity to spearhead this new model.

We'll provide updates in future blog posts as well as on the mailing lists of the project involved as we achieve significant progress toward the goals of the grants.

These are exciting times, funding for core PyData projects is accelerating.