This guest post is written by David Masad. David is a PhD candidate at George Mason University’s Department of Computational Social Science, where he studies international conflict and cooperation using agent-based modeling, event data, and network analysis. You can follow him on Twitter at @badnetworker.
'SciPy' has a few different meanings. It is a particular Python package, which brings together fast, efficient implementations of many key functions and algorithms for scientific computation. It's also the label for the broader scientific Python stack, the set of libraries and tools that make Python an increasingly popular language for science and research. Finally, it's what everyone calls what's nominally the Scientific Computing with Python conference, which for the past few years has been held every summer in Austin, TX.
This year, it involved two days of intensive tutorials; three days of presentations, talks, and discussion sections; and two more days of informal coding 'sprints'. Though I've been using the scientific Python tools for several years now, this was my first time attending SciPy. I even got a little "1st SciPy" sticker to add to my conference badge. For about five days, I got to be a part of a great community, and experience more Python, brisket and Tex-Mex than I realized was possible.
Part of the fun of this particular conference was the opportunity to talk to researchers working in areas far afield from my own. For example, over lunch burritos I got to talk shop with someone working at a climate research firm, and discovered interesting overlaps between weather and political forecasting. An astronomy talk contained some great insights into building community consensus around a common software tool. And a talk that was officially about oceanography had some very important advice on choosing a color palette for data visualization. (If you followed the links above, you saw that all the SciPy talks are available online, thanks to Enthought's generous sponsorship -- it's not too late to see any SciPy talks that seem interesting to you.)
The SciPy attendees from the DC area were a good cross-section of the diverse scientific Python community in general. There were an epidemiologist and a geneticist, a few astronomers, a government researcher, a library technologist, and a couple of social scientists (myself included). (If there were any DC-area geo-scientists there, I didn't get a chance to meet them).
There was also plenty of material directly applicable to data science. Chris Wiggins, the chief data scientist at the New York Times, gave the first day's keynote, with plenty of good insight into bringing data science into a large, established organization. Chris Fonnesbeck gave an engaging talk on the importance of statistical rigor in data science. Quite a few of the presentations introduced tools that data scientists can install and use right now. These include Dask, for out-of-core computation; xray, an exciting new library for multidimensional data; and two talks on using Docker for reproducible research. There was a whole track devoted to visualization, including a talk on VisPy, a GPU-accelerated visualization library, that gave the conference one of its big 'Wow!' moments. And the future of Jupyter (still better known as the IPython Notebook) was announced in a 5-minute lightning talk, between demos of bad-idea ways to call Assembly directly from Python and Notebook flight sim widgets (which Erik Tollerud immediately dubbed 'planes on a snake').
Not only did I get to learn a lot from other people's research and tools, I got to present my own. Jackie Kazil and I unveiled Mesa, an agent-based modeling framework we've been building with other contributors. The sprint schedule after the conference proper gave us a chance to work with new collaborators who discovered the package at our talk the day before. A couple extra heads, and a couple days of extra work, mean that Mesa came out of SciPy noticeably better than when it came in. Quite a few other tools came out of the sprints with some improvements, including ones at the core of the scientific Python stack. Getting to work beside (and, later, drink beer with) such experienced developers was an educational opportunity in itself.
SciPy isn't just for physical scientists or hardcore programmers. If you use Python to analyze data or build models, or think you might want to, you should absolutely consider SciPy next year. The tutorials at the beginning can help novices to experts learn something new, and the sprints provide an opportunity to gain experience putting that knowledge to work. In between, the conference provides a great opportunity to gain exposure to a great variety of Python tools, and the community that builds, maintains and uses them. And even if you can't attend -- the videos of the talks are always online.