Reflecting on the NBA Hackathon

Apr 22, 2015; Memphis, TN, USA; A detailed view of basketballs in front of the NBA Playoffs logo at the FedexForum before the game between the Memphis Grizzlies and the Portland Trail Blazers in game two of the first round of the NBA Playoffs. Mandatory Credit: Justin Ford-USA TODAY Sports
Apr 22, 2015; Memphis, TN, USA; A detailed view of basketballs in front of the NBA Playoffs logo at the FedexForum before the game between the Memphis Grizzlies and the Portland Trail Blazers in game two of the first round of the NBA Playoffs. Mandatory Credit: Justin Ford-USA TODAY Sports /
facebooktwitterreddit

This past Saturday, Senthil and Chris both had the opportunity to compete in the inaugural NBA Hackathon event hosted by the NBA at Terminal 23 in New York City. For the Nylon Calculus group, this event also marked the debut of Team Nylon who competed against over 200 students representing over 50 different universities across the country. While Team Nylon walked away with a second place finish, they also left the event with memorable experiences, thoughts and lessons learned that they will carry forward into their future work. Here are their thoughts. . .

Chris Pickard: First off, before I jump into some of my immediate takeaways, I want to take a moment or two to give a huge thanks to the NBA, the NBA Stats department and all of the NBA team representatives and writers that were at the event. I have participated, organized and run first-time events, and, as far as first-time events go, the Hackathon Saturday was well done. As with any first-time event, there were some areas that could be improved, but I look at that as an exciting prospect considering how well the event went. Certainly, a big thanks to everyone who was apart of it and I look forward to seeing how this event grows in the future!

Heading into the event, I really had no idea what to expect. I had plenty of familiarity with the idea of hackathons because, quite frankly, it’s hard not escape them out here in Silicon Valley, but I had never personally competed in one. I had no idea what the level of competition was going to be or how the final prompts were going to ultimately shake out. On top of it all, this was the first time that Senthil and I had ever met, let alone worked together in a time-constrained event.

More from Nylon Calculus

Senthil Natarajan: Absolutely second that thanks. This was a great event, and the NBA definitely has something promising on its hands. It was worth the trip out to New York for sure, and the atmosphere was incredible, from being in Terminal 23 to all the NBA personnel who were there. I’ve never been much of a natural coder or software guy, but am an absolute basketball junkie. So when Chris reached out about this event, I figured there was no way I wasn’t going to go to NYC. I will say that while we sat at Rockefeller Center eating Chipotle, we claimed upfront that the experience was really what counted. However, both of us had enough of a competitive streak to implicitly acknowledge we were going to go full throttle for the grand prize.

And well, certainly not bad for our first hackathon ever, cuz. I’m a little sad that I didn’t make time for Steak ‘n’ Shake, but I did see enough Halal Guys carts to fill any longings I may have had.

Chris Pickard: While I would love to get into the stories, funny moments and the overall adventure of the event, I want to take a step back and reflect on our 8+ hour sprint to produce our hack with an additional final presentation and highlight some of the major takeaways we had from the event.

Senthil: Let’s get to it then.

The importance of “talking” basketball:

Chris: From the start of the event, it was clear that Senthil and I weren’t necessarily the most knowledgeable about hard statistics or could code in as many languages as some of the other competitors. That’s not to discredit our own skills, but Senthil (Electrical Engineering major) and I (Civil Engineering major) just don’t necessarily come from stereotypical analytics backgrounds. However, what we do know well is the game of basketball. I stress game because the game itself is complex and in a constant state of change between seasons, games, and possessions.

Before someone can build a model, manipulate data and spit out some meaningful numbers about basketball, you have to know how they game is played as well as what it means within the flow of a game. For example, when we decided to work on measuring defensive versatility for our hack, our initial discussions had nothing to do with models, code or math, but what versatility meant on defense. “Why was Draymond Green valued as such a versatile defender?,” was a question we asked and we discussed several hypothetical in-game situations where defensive versatility shows up, but also, more importantly, how it shows up.

Senthil: Versatility isn’t truly based on classical positional assignments because guarding Tyson Chandler and Channing Frye are much different even though they are classified as centers. As a result, it made sense that defensive versatility is intertwined with the offensive player’s style and we should probably consider using some type of clustering model to paint a better picture of what these various scoring styles look like. We limited “performance” to shot events, mainly due to the scope of the event, but understood that this would be a factor of where the shooter and the defender where on the court.

Keeping things grounded in basketball helped us understand that, in this case, we didn’t necessarily need to create a new complex system to gain insight. That’s why when we presented our work in the final round, we excluded the final scores themselves, and we presented how players fared defending against each type of scoring style, rather than just the final versatility numbers. Because really, those numbers in a vacuum don’t really mean anything. What does matter is how one player’s scores compare to that of others’, the context surrounding that, and the implications of that score. This is especially critical for defense, where a team’s defensive scheme and the ability of surrounding players can factor so much into evaluating individual performance. This leads us nicely into the next point…

Recognizing the complexity of basketball

Chris: One thing that Senthil and I noticed during the event and while we listened to other presentations was how often the complexity of basketball, especially in the NBA, was not acknowledged. This isn’t to slight any of the competition, but there were often times that various presentations could have been elevated had there been recognition of the many factors that can affect a possession, game or season. For us, this showed up in two main ways while we put our project together.

First, to identify defender-shooter matchups we used the nearest defender at the moment of the shot, but this doesn’t mean that defender was the primary defender. Imagine a situation when Channing Frye is put in a pick-and-pop, something he did a lot in Phoenix when he was with Goran Dragic. In these cases, his nearest defender was typically not his primary defender, which shows a limitation in the data we used for our model, but with more granulized SportVU data we could eventually account for strengthening our model’s results.

unnamed
unnamed /

Second, after setting up and running the model, we found James Harden to be the best defender at guarding perimeter specialists, which at first seemed like a major red flag, until you start understanding how the Rockets try to use Harden on defense. In terms of defensive schemes, the Rockets often moved Harden to guard off-the-ball perimeter specialists (saving him for offense) and, as a result, will start most defensive possessions in the corners.

In addition, many of Harden’s actual primary defensive matchups result with the offensive player blowing by Harden, alternating the identity of the nearest defender away from Harden, which would reduce the number of times he “guarded” a certain style of player. By taking into account some of the finer aspects of a basketball game, we were able to get a better understanding of why our results turned out a certain way. Sometimes it elevated our insight and other times it pointed to things we or the data did not consider, which opens up the door for a better model in the future. It was something we and other people we talked to didn’t see more of during the event.

Senthil: I think there’s a natural tendency to get lost in how cool your work is or how your numbers look that sometimes you lose a little perspective about what the actual point of your project is, what you’re actually attempting to learn. And this happens to everyone, us included. A little introspection and self-criticism can always help smooth out the irregularities in our work or deliver key insights.

My first (pretty embarrassing) lesson with this was actually when I was putting together some radar charts for characterizing offense. I was so enamored by those final charts that I failed to notice one detail for several days: there was a quirk in Andre Drummond’s radar chart in the FT% field that made it look like he was Ray Allen at the charity stripe. Well, upon further introspection, it turned out that when I had normalized all the values, Drummond was actually so historically bad that it had just broken the chart altogether, a feature that I had to go back and account for. And that kind of realization can only happen if you’re willing to step back and come at analytics from a basketball perspective as well as a math or statistics perspective.

Analytics is a tool to augment, not replace classical basketball analysis, and that fact that you need both to work hand in hand was definitely something that the judges and other front office personnel placed an emphasis on in our conversations. Context matters just as much as concepts. With respect to our own project, it opened up the key realizations for exactly where our model still needs to improve and at each step of the way, it was important for us to ask, “Are we solving for what we meant to solve for? What exactly is this line of code doing for us? Where are the dependencies?” In brainstorming what data sets we wanted to use (rather than falling into the trap of trying to do too much), we had to first ask, “If we were playing defense or if we were coaching defense, what are the key markers that we’d be looking for?”

The opportunity to network:

Chris: From our perspective, one of the coolest parts of the event was the representation of about a dozen or so teams along with NBA league office members and well-respected basketball writers such as Zach Lowe that were in attendance. While from a pure basketball analytics perspective, the actual hackathon competition was great, even if it was stressful at times for just the two of us to complete our entire project in the allotted time. Being able to meet and talk to guys in the industry was special.

Truth be told, these guys are pretty normal and cool people who love basketball and provided a lot of insight to it and analytics whether it was on our project or ideas for future work. They were just as curious about basketball as we were and that made for some interesting conversations whether it was with Ken Pomeroy or Zach Lowe. It was also something Senthil and I were surprised more more participants weren’t doing as at times it felt like we were the only ones mingling.

Senthil: Chris conducted a masterclass in networking that deserves its own space, but I digress. From my experience, I will say that it takes a certain mindset to go and cold-talk to people in various positions of authority. The opportunity was there, but it’s hard to break out of a comfort zone. As such even when there was time after the hacks were submitted, participants were content to rather play ping pong or 2K than approach someone completely strange or to break into an existing conversation with people whom they were totally unfamiliar with. While Chris or I can hold our own in any conversation, it definitely added credibility and ease to be able to have Seth Partnow introduce us.

Next: Introducing WNBA RAPM

One great idea we heard from a team representative was to assign each hackathon team a mentor who works for an NBA team or within the league. Then by instituting a halftime break, each team would then meet with their mentor, forcing interaction between the competitors and team representatives, but would also allow hackathon teams the opportunity to get feedback with a much broader and in-depth context of the basketball implications in their hacks. That’s actually how we were able to more strongly consider the impact of team defensive scheme on our final output. Hopefully, as the event evolves this issue is addressed more. This is one of the great strengths of this Hackathon, and it would be a shame to see it not properly taken advantage of.