Sloan 2014: Putting Player Tracking to Work

Sloan Conference 2014

The honeymoon is over, if it wasn’t already. The marriage of basketball and modern analytics had long ago been consummated, but these past three or four years have felt like something of an ongoing celebration with the advent of player tracking technologies provided by STATS LLC. Bursting onto the scene just as the MIT Sloan Sports Analytics Conference was growing into a bigger audience and larger venues, the advent of the SportVU camera system earned the ol’ Keanu ‘whoa’ from the hoops community. In the years that followed, all the exploratory work done with the new data came with a caveat: in order to begin approaching truly accurate conclusions, the cameras had to find their way into every arena.

Now wherever there is NBA basketball, you can find player tracking. The cameras have been adopted league-wide, thanks in part to the diligent work of STATS’ Brian Kopp and his team, and the dataset has begun growing in earnest. As we near the end of the first full season since SportVU assimilation, the shine is slowly starting to come off the new relationship. That’s a good thing.

Because now it’s time to get to work.

If the speaking panels at Sloan have become occasionally stale over the years – though ‘The Science of the Deal,’ in which Houston Rockets GM Darryl Morey and Golden State Warriors’ GM Bob Meyers discussed the Dwight Howard and Andre Iguodala deals, was a pleasant surprise – the research papers are approaching lunch-pail status. As vast as the possibilities are with SportVU, many have yet to come to fruition. Pick-and-rolls have been particularly tricky to capture, and while STATS has internally been developing a proprietary algorithm a team from MIT helmed by Arman McQueen, Jenna Wiens and John Guttag took up finding a solution with their paper, titled ‘Automatically Recognizing On-Ball Screens’.

For immediate public purposes, the merits of the paper are almost beside the point. You don’t need to be told how to recognize an on-ball screen. You can already do that in real time. The work is merely filling in the gaps for what is still a very new technology. Computers need to be able to see, too. It might have come off dry in a presentation format, but important work is sometimes ugly. Functionality is often developed in the trenches. Whether it’s more functional than anything STATS has developed remains to be seen.

If the group from MIT was focused more on SportVU in practice, a group from Harvard University – Dan Cervone, Alexander D’Amour, Luke Bornn and Kirk Goldsberry – traded function for theory with ‘POINTWISE: Predicting Points and Valuing Decisions in Real Time with NBA Optical Tracking Data’. The idea was, simply, to assign a value to all the actions that are and aren’t made on the basketball court. What is the value of LeBron James taking a three, for instance, versus driving down the left side of the paint or firing a cross-court pass to the weakside corner? How many potential points is James giving up once he makes the decision and the other possibilities disappear into the ether?

It’s an idea that has been kicked around the basketball analytics community for years, but this is as close as anyone has publicly come to creating an operational model. Getting closer doesn’t take us out of the theoretical realm, however, and we’re still a ways off from the idea becoming a practical reality – in part due to the massive processing requirements of such an endeavor.

Bridging the gap between the aforementioned projects was the group from California, Second Spectrum. Headed by Rajiv Maheswaran, Second Spectrum came onto the Sloan scene a few years ago using the SportVU data to track rebounds. This year, they tried to determine who the best rebounders in the NBA are by breaking up the timeline of a rebound – from the shot to the miss to the conversion of the rebound itself – and in turn assigning values to hustle, positioning and conversion rate. You can read about the results here, but in order to do any of this the team has some very practical tools put into place. Namely, they can show you, with visuals, where a rebound is most likely to fall based on where the shot is taken.

Even if it’s telling players things they already know, that confirmation has value. With an image, Second Spectrum can confirm a lifetime of strategy and experience. Well, in time at least. We aren’t drawing any absolute truths from SportVU until we have years of historical data to work with, but if time is your only opponent, if you’ve already developed the necessary tool, then you’ve provided a valuable service.

As much fun as it is to be wowed, as limitless as the possibilities are, the data eventually needs to be put to work. Every team in the league, to some degree, is trying to figure out how to do that behind the scenes. But those teams aren’t telling their secrets. They have no reason to. Most of the public research being done is showed off at Sloan – which offers a $20,000 prize to the best paper – and even some of those groups are confined to working with year-old data from when the SportVU cameras weren’t in every arena.

SportVU made a very public entrance at Sloan years ago, but now that STATS has partnered with the NBA it is a very private enterprise. You’ll see the available data on NBA.com/Stats grow and grow – pick-and-rolls and rebounding locations could be eventually be added – but it will always be a glimpse of what is really possible. We’ll still see good ideas at Sloan during the first weekend of every March, but the party is over and the cleanup crew has arrived. Everyone has taken the data home and shut the door. It’s time to work now.