Skip to content

At Strata, “hardcore” data science is pretty fluffy

Last week, I attended O’Reilly’s data-centric conference, Strata. It’s my fourth Strata and my third on the west coast, so I’m starting to get a pretty good feel for the show’s evolution over time and some of the contrast across coasts as well.

I keep going into the show with high expectations for the “Hardcore Data Science” track (“Deep Data” in 2012 [writeup]), which is framed essentially as continuing education for professional data scientists. Unfortunately, both years I’ve attended, it fell tragically short of that goal of educating data scientists. In 2012, I sat through the whole day and heard 2–3 talks where I learned something new, but this time I was so disappointed that I left around 11am in favor of the data-driven business track. In talking to other attendees, general reception was that the Google talk on deep learning was great, but the rest of the day was a disappointment in terms of technical content and learning practical and usable techniques.

I must admit I’m deeply surprised that O’Reilly didn’t get negative feedback last time around that it should’ve applied to this year’s “hardcore” program, as I consider the company among the top couple of professional conference organizers around, across the widest set of topics.

One of the challenges with Strata is catering to a diverse set of audiences, and O’Reilly’s done an excellent job with the “I want to learn Hadoop / new Big Data tech X” crowd. More recently, they’ve also done very well reaching out to the business-level audience trying to learn about the value of data. However, it seems like the technical core of the conference is gradually being left in the lurch in terms of their opportunity to learn, although there’s always the hallway track and whatever marketing value they get out of their talks and tutorials.

I would suggest that the intensive data-science track at future Strata conferences be made much more technical, that the talks become sufficiently practical that they’re handing out links to GitHub or other real-world, low-level implementations at the end of the talk, and that this shift in topic and intended audience be very clearly communicated to the speakers. Other than that slip-up, good show — I heard good things about the tutorials, the business day, and talks for the mid-level audience throughout the rest of the conference.


Categories: big-data, data-science.