Written in Python, scikit-learn is an open source machine learning software package used widely across the natural and social sciences (the “software paper” that introduced scikit-learn in 2011 has been cited over 4,700 times). Its maintainers have identified a set of improvements that would make it substantially more efficient for scientific users and enable more reproducible research, but which would require more focused time than any contributor can currently offer. This grant provides funds to Columbia University’s Andreas Mьller, one of the current core maintainers of scikit-learn, to design and implement the identified improvements. These include more flexible data types, better integration with Jupyter notebooks for model exploration, and some technical fixes that will substantially improve platform stability and performance.

A whole lot of science takes place in spreadsheets. Many researchers still bring their data into Excel as a convenient environment for exploration and analysis. Unfortunately, Excel has none of the attributes of a modern platform for reproducible computational research: it is not easily extensible to interoperate with data repositories; does not easily allow for version control; and cannot take advantage of substantial investments in open source scientific software packages. Nokome Bentley, a New Zealand-based fisheries scientist and software developer, has been developing a project called Stencila Sheets, an authoring tool that offers users familiar Google Docs–style interfaces, but is something quite different under the hood. His vision is a spreadsheet where each cell can hold data or code written in R, Python, Julia, or several other computing languages, with the output of a given cell addressable by any other cell in the sheet. The proximate goal is not to develop a direct competitor to Excel, but rather to offer spreadsheet users an easy bridge into the open-source ecosystem of reproducible computational science. Funds from this grant will allow further development of the Stencila platform over the next year, including increased integration with the Jupyter computing ecosystem, the development of a standalone desktop client, and the addition of features like real-time collaboration and import/export from other platforms.

As computers and computational analysis becomes an increasingly central part of scientific practice, more and more scientists are becoming better and better at writing and amending software and code. What scientists often don’t know how to do, however, is to transition a piece of software from something built in their own lab to a sustainable open source, community-driven project. Open source software development, however, has proven to be one of the singularly most influential paths to widespread adoption, dissemination, and innovation in software development. In order for open source to be a viable sustainability strategy for some scientific software, there needs to be better support and training for scientists to “do open source.” This grant funds an initiative at the Mozilla Foundation to help train scientists in the launch and management of open source software development projects. Funded activities include the development of an expanded open science curriculum that details best practices for open source software development, project management, community organizing and facilitation, engaging noncoders, and data management. Additional grant funds support a series of workshops, online chats, and conference calls on these and related topics and and a community-based mentorship program.

In 2013, the Foundation partnered with the Gordon and Betty Moore Foundation to launch a five-year, $37.8 million initiative that aspired to advance data-intensive scientific discovery, empowering researchers to be vastly more effective by utilizing new methods, new tools, new partnerships, and new career paths. The initiative led to the funding of three university partnerships, one with New York University, one with the University of California, Berkeley, and one with the University of Washington, to create Data Science Environments (DSEs) that would innovate new models for advancing data science at American universities. The centers would focus on three core goals: crafting meaningful interactions between data scientists and disciplinary scientists, experimenting with long-term, sustainable career paths for data scientists in the university system, and developing new analytical tools and research practices that will empower scholars to work effectively with data. Funds from this grant support a team at Abt Associates to document and evaluate the individual and joint progress of the three Moore-Sloan Data Science Environments. Combining qualitative and quantitative data collection and analysis, the Abt team will document DSE goals and activities, provide annual reports to each DSE on its progress, and produce three major reports: a landscape survey of data science efforts in top U.S. research universities broadly (to contextualize the DSE activities); an implementation study of the actual execution of the DSE activities at the three universities; and an impact study that aims to understand the consequences of the unique DSE interventions on individual career paths and research outcomes as well as on institutional structures.