Deep dive into Project Tungsten: Bring Spark closer to bare metal
Project Tungsten focuses on substantially improving the efficiency of memory and CPU for Spark applications, to push performance closer to the limits of modern hardware. This effort includes three initiatives:
Memory Management and Binary Processing: leveraging application semantics to manage memory explicitly and eliminate the overhead of JVM object model and garbage collection Cache-aware computation: algorithms and data structures to exploit memory hierarchy Code generation: using code generation to exploit modern compilers and CPUs
Project Tungsten will be the largest change to Spark’s execution engine since the project’s inception. In this talk, we will give an update on its progress and dive into some of the technical challenges we are solving.
[A version of this articles appears on the O’Reilly Radar.]
Subscribe to the O’Reilly Data Show Podcast to explore the opportunities and techniques driving big data and data science.
As tools for advanced analytics become more accessible, data scientist’s roles will evolve. Most media stories emphasize a need for expertise in algorithms and quantitative techniques (machine learning, statistics, probability), and yet the reality is that expertise in advanced algorithms is just one aspect of industrial data science.
During the latest episode of the O’Reilly Data Show podcast, I sat down with Alice Zheng, one of Strata + Hadoop World’s most popular speakers. She has a gift for explaining complex topics to a broad audience, through presentations and in writing. We talked about her background, techniques for evaluating machine learning models, how much math data scientists need to know, and the art of interacting with business users.
Making machine learning accessible
People who work at getting analytics adopted and deployed learn early on the importance of working with domain/business experts. As excited as I am about the growing number of tools that open up analytics to business users, the interplay between data experts (data scientists, data engineers) and domain experts remains important. In fact, human-in-the-loop systems are being used in many critical data pipelines. Zheng recounts her experience working with business analysts:
It’s not enough to tell someone, “This is done by boosted decision trees, and that’s the best classification algorithm, so just trust me, it works.” As a builder of these applications, you need to understand what the algorithm is doing in order to make it better. As a user who ultimately consumes the results, it can be really frustrating to not understand how they were produced. When we worked with analysts in Windows or in Bing, we were analyzing computer system logs. That’s very difficult for a human being to understand. We definitely had to work with the experts who understood the semantics of the logs in order to make progress. They had to understand what the machine learning algorithms were doing in order to provide useful feedback. Continue reading “Bridging the divide: Business users and machine learning experts”
[A version of this article appears on the O’Reilly Radar.]
One of my favorite books from the last few years is David Epstein’s engaging tour through sports science using examples and stories from a wide variety of athletic endeavors. Epstein draws on examples from individual sports (including track and field, winter sports) and major U.S. team sports (baseball, basketball, and American football), and uses the latest research to explain how data and science are being used to improve athletic performance.
In a recent episode of the O’Reilly Data Show Podcast, I spoke with Epstein about his book, data science and sports, and his recent series of articles detailing suspicious practices at one of the world’s premier track and field training programs (the Oregon Project).
Nature/nurture and hardware/software
Epstein’s book contains examples of sports where athletes with certain physical attributes start off with an advantage. In relation to that, we discussed feature selection and feature engineering — the relative importance of factors like training methods, technique, genes, equipment, and diet — topics which Epstein has written about and studied extensively:
One of the most important findings in sports genetics is that your ability to improve with respect to a certain training program is mediated by your genes, so it’s really important to find the kind of training program that’s best tailored to your physiology. … The skills it takes for team sports, these perceptual skills, nobody is born with those. Those are completely software, to use the computer analogy. But it turns out that once the software is downloaded, it’s like a computer. While your hardware doesn’t do anything alone without software, once you’ve got the software, the hardware actually makes a lot of a difference in how good of an operating machine you have. It can be obscured when people don’t study it correctly, which is why I took on some of the 10,000 hours stuff.
… You might think Usain Bolt moves his legs fast, but he actually repositions his legs at the same rate as your grandmother when she’s running as fast as she can, or maybe your mother if your grandmother is a little older. Sprinters don’t win by moving their legs through the air faster. They win by putting five times their body weight into the ground as fast as humanly possible. … Literally, sprinting is limited by the contractile speed of the muscle fibers, so you need a lot of those fast twitch muscle fibers. … There’s a lot of longitudinal data with tens of thousands of people who are tracked longitudinally that show that, whether you like it or not, slow kids do not become fast adults. … Speed is slightly predictable in a broad sense.
Subscribe to the O’Reilly Data Show Podcast
As in finance and other domains, innovations only remain proprietary for a limited amount of time. For one thing, athletes, trainers, and sports scientists bounce around between organizations and bring ideas along with them, but also breakthroughs can be observed and reverse-engineered. To some extent, this means that athletes and teams start looking alike (as I lamented in our conversation, there is a trend toward hyper-specialization — for example, most NBA teams employ players adept at shooting “corner threes”). Epstein cited a recent example:
There’s a really funny example of that happening in a sport called skeleton, which is one of those new sports where innovation makes a huge impact. It’s a winter sport where people slide face-first down an icy track. … Everyone used to use two hands on the sled, then you run with it and you jump on it. This [British] coach was worried that the Americans had better equipment and were going to destroy his team. … These [British] guys basically invented the one-hand start. They had been training a certain way for several years. He gave them two hours to just go be creative, whatever. Do something stupid. They come back asking, ‘Is it within the rules to do it one-handed?’ He looked; it’s not against the rules. They keep it secret, and when they broke it out, they broke the start world records left and right. Then everybody started using it right away, so it literally overnight transformed what everyone does in this sport.
Pattern recognition using estimates and “cheating”
As a longtime Tour de France fan, I’ve noticed that a group of fans and longtime watchers have taken to estimating various factors, like power output. Oftentimes, they compare an array of metrics that riders have generated in recent editions of the tour to similar metrics from the “doping era.”
Teams and officials previously labeled their efforts as pseudoscience, only to backtrack when it turned out that their power output estimates were extremely accurate. Using their own data and calculations, cycling fans are, in essence, using comparative and longitudinal studies to flag suspicious performance numbers. This has led to calls for teams and riders to provide more transparency by supplementing biological passportswith the release of “training and racing log files.”
As Epstein noted, this type of comparative data would be insufficient in a court case, and in many situations good old-fashioned investigative journalism (sources and leaks) is what ultimately exposes cheaters. Nevertheless, it’s still good to see cycling fans engage with and pressure teams and race organizers, to release more data:
In those past eras of doping, like in the Lance Armstrong era, you look at what happens when the EPO test comes in and suddenly power outputs plummet. This year, in some cases, they look like they’re back to where they were after that. It’s not like guys had stopped training hard; they stopped doing EPOs. That sport for sure has earned the suspicion it gets. We have to be careful because the bicycles are improving, the weather changes — there are a lot of variables. But with the history of the sport and the fact that they were calling measurements that turned out to be quite accurate ‘pseudoscience,’ I think if they’re complaining about people being gadflies, that’s crazy. … It’s truly interesting, too, which sports the fans and enthusiasts engage with in that way.
Epstein pointed out that cheating can turn off fans, and it also makes comparative and longitudinal studies difficult to do:
People used to say women are going to catch up on men when they have more opportunity, but actually men are pulling away now. The gap is widening. I think it’s partly because a lot of the women’s records are stuck. Steroids, which are just testosterone analogues, have a much greater effect in women than they do in men. … We know in the past, there was this era of mega-doping. All these documents have now come out related to East Germany, and there was this very systematic, enormous amount of doping, so tons of women’s records are stuck and nobody even gets close to them most of the time. It’s a bummer.