Making data tell its story
A picture may or may not be worth a thousand words, but a picture is certainly worth a thousand numbers. The problem with most data analysis algorithms is that they generate a set of numbers. To understand what the numbers mean, the stories they are really telling, you need to generate a graph. Edward Tufte’s Visual Display of Quantitative Information is the classic for data visualization, and a foundational text for anyone practicing data science. But that’s not really what concerns us here. Visualization is crucial to each stage of the data scientist. According to Martin Wattenberg (@wattenberg, founder of Flowing Media), visualization is key to data conditioning: if you want to find out just how bad your data is, try plotting it. Visualization is also frequently the first step in analysis. Hilary Mason says that when she gets a new data set, she starts by making a dozen or more scatter plots, trying to get a sense of what might be interesting. Once you’ve gotten some hints at what the data might be saying, you can follow it up with more detailed analysis.
There are many packages for plotting and presenting data. GnuPlot is very effective; R incorporates a fairly comprehensive graphics package; Ben Fry’s Processing is the state of the art, particularly if you need to create animations that show how things change over time. At IBM’s Many Eyes, many of the visualizations are full-fledged interactive applications.
Nathan Yau’s FlowingData blog is a great place to look for creative visualizations. One of my favorites is this animation of the growth of Walmart over time. And this is one place where “art” comes in: not just the aesthetics of the visualization itself, but how you understand it. Does it look like the spread of cancer throughout a body? Or the spread of a flu virus through a population? Making data tell its story isn’t just a matter of presenting results; it involves making connections, then going back to other data sources to verify them. Does a successful retail chain spread like an epidemic, and if so, does that give us new insights into how economies work? That’s not a question we could even have asked a few years ago. There was insufficient computing power, the data was all locked up in proprietary sources, and the tools for working with the data were insufficient. It’s the kind of question we now ask routinely.
Data science requires skills ranging from traditional computer science to mathematics to art. Describing the data science group he put together at Facebook (possibly the first data science group at a consumer-oriented web property), Jeff Hammerbacher said:
… on any given day, a team member could author a multistage processing pipeline in Python, design a hypothesis test, perform a regression analysis over data samples with R, design and implement an algorithm for some data-intensive product or service in Hadoop, or communicate the results of our analyses to other members of the organization 3
Where do you find the people this versatile? According to DJ Patil, chief scientist at LinkedIn (@dpatil), the best data scientists tend to be “hard scientists,” particularly physicists, rather than computer science majors. Physicists have a strong mathematical background, computing skills, and come from a discipline in which survival depends on getting the most from the data. They have to think about the big picture, the big problem. When you’ve just spent a lot of grant money generating data, you can’t just throw the data out if it isn’t as clean as you’d like. You have to make it tell its story. You need some creativity for when the story the data is telling isn’t what you think it’s telling.
Scientists also know how to break large problems up into smaller problems. Patil described the process of creating the group recommendation feature at LinkedIn. It would have been easy to turn this into a high-ceremony development project that would take thousands of hours of developer time, plus thousands of hours of computing time to do massive correlations across LinkedIn’s membership. But the process worked quite differently: it started out with a relatively small, simple program that looked at members’ profiles and made recommendations accordingly. Asking things like, did you go to Cornell? Then you might like to join the Cornell Alumni group. It then branched out incrementally. In addition to looking at profiles, LinkedIn’s data scientists started looking at events that members attended. Then at books members had in their libraries. The result was a valuable data product that analyzed a huge database — but it was never conceived as such. It started small, and added value iteratively. It was an agile, flexible process that built toward its goal incrementally, rather than tackling a huge mountain of data all at once.
This is the heart of what Patil calls “data jiujitsu” — using smaller auxiliary problems to solve a large, difficult problem that appears intractable. CDDB is a great example of data jiujitsu: identifying music by analyzing an audio stream directly is a very difficult problem (though not unsolvable — see midomi, for example). But the CDDB staff used data creatively to solve a much more tractable problem that gave them the same result. Computing a signature based on track lengths, and then looking up that signature in a database, is trivially simple.
Hiring trends for data science
It’s not easy to get a handle on jobs in data science. However, data from O’Reilly Research shows a steady year-over-year increase in Hadoop and Cassandra job listings, which are good proxies for the “data science” market as a whole. This graph shows the increase in Cassandra jobs, and the companies listing Cassandra positions, over time.
Entrepreneurship is another piece of the puzzle. Patil’s first flippant answer to “what kind of person are you looking for when you hire a data scientist?” was “someone you would start a company with.” That’s an important insight: we’re entering the era of products that are built on data. We don’t yet know what those products are, but we do know that the winners will be the people, and the companies, that find those products. Hilary Mason came to the same conclusion. hHer job as scientist at bit.ly is really to investigate the data that bit.ly is generating, and find out how to build interesting products from it. No one in the nascent data industry is trying to build the 2012 Nissan Stanza or Office 2015; they’re all trying to find new products. In addition to being physicists, mathematicians, programmers, and artists, they’re entrepreneurs.
Data scientists combine entrepreneurship with patience, the willingness to build data products incrementally, the ability to explore, and the ability to iterate over a solution. They are inherently interdiscplinary. They can tackle all aspects of a problem, from initial data collection and data conditioning to drawing conclusions. They can think outside the box to come up with new ways to view the problem, or to work with very broadly defined problems: “here’s a lot of data, what can you make from it?”
The future belongs to the companies who figure out how to collect and use data successfully. Google, Amazon, Facebook, and LinkedIn have all tapped into their datastreams and made that the core of their success. They were the vanguard, but newer companies like bit.ly are following their path. Whether it’s mining your personal biology, building maps from the shared experience of millions of travellers, or studying the URLs that people pass to others, the next generation of successful businesses will be built around data. The part of Hal Varian’s quote that nobody remembers says it all:
The ability to take data — to be able to understand it, to process it, to extract value from it, to visualize it, to communicate it — that’s going to be a hugely important skill in the next decades.