During the Williams get-together at the March Meeting last week, somebody mentioned that they had been skimming through sessions and noticed a talk about numerical modeling of basketball. I wondered aloud whether it was by Brian Skinner who actually has written articles about basketball analysis (the most recent is here on the arxiv), but they didn’t remember. I looked it up later, and it was by a grad student at Cornell, and was part of a session on “Social Interactions” to be held a day or so later. It also got written up in a press release from Cornell, which Sabine Hossenfelder retweeted later.
I didn’t have anything else that I desperately wanted to see in that time block, so I went to check it out. The basic idea was to use the rich dataset of player position tracking data that the NBA collects on every game these days, and create a function that assigns a value to each position on the floor based on the position of other players and the ball. This in principle allows you to create a map of the optimal position for any selected player in a particular situation, which you can then compare to where the player actually was.
This lends itself to the making of nifty animations, but both Brian (who also attended the talk) and I thought its analysis was a little wonky in basketball terms. The player the speaker highlighted was defending a man who was spotted up in the wing/corner area, and the model wanted them to be further away, leaving the corner three wide open. That’s not a great idea in the modern NBA, to say the least, but it could maybe be tweaked to fix that.
The general idea of assigning values to positions, though, combined with the general conference atmosphere, led me to think of an idea along these lines that might be of direct use to the APS community. I’m not really the right guy to do it, though, so I’ll just throw it out here on Substack, and somebody else can do the numerical work.
The problem to be addressed is the problem of getting food at the March Meeting in an optimal way, given the constraints of session schedules and restaurant locations. This is, obviously, a problem that was being solved in real time, every day, by the attendees of the conference that just happened, but having spent a few days looking at the results, I’m pretty sure they were very far from the optimal solution.
If you look at the map above, you can see the general area of the Caesar’s Forum conference center (at the upper right), which is directly connected to the Harrah’s casino, and right next to a pedestrian street (the “LINQ Promenade”) lined with restaurants and bars— this runs from the Yard House past the Gordon Ramsey Fish and Chips place and In-n-Out Burger to Maxie’s diner. And of course, it’s right in the middle of the Las Vegas Strip, with casinos galore.
In the peak lunch time windows— 11am to 1pm— physicists expanded out from the conference center, conveniently tagged with their APS meeting badges, which basically nobody took off, ever (a real divergence from meeting etiquette in my day…). They didn’t diffuse all that far, though. Maxie’s was a mob scene, and there were massive lines at the In-n-Out and the “sushi burrito” place across the street, and likewise the food court area in Harrah’s. The crowd dropped off very steeply, though— the Ramsey fish and chips stand was busy, but not as slammed as In-n-Out, and you could readily walk in at any of the sit-down places up the street. The food court area of the Flamingo, basically indistinguishable from that in Harrah’s in terms of food on offer, was a ghost town.
Now, it’s true that these places were a bit farther away from the conference center, but to my mind that would be more than made up for by the absence of a long line. Which is why I say the solution was sub-optimal— waiting fifteen minutes for a slice of pizza in the Harrah’s food court is less good than walking an additional five minutes to the Flamingo’s food court, where you could get the exact same Sbarro-ass pizza with no line (as I did one of the days I was there).
So, I put this out there for some enterprising grad student with some numerical modeling chops. I’m pretty sure you could construct a density function that would assign values to positions based on distance from the conference center, expected wait time, and price, and come up with an optimal location for lunch on a given day. The model could later be extended to include cuisine preferences, though this might require the use of gauge theory.
Las Vegas, with its unusually high density of eateries, would’ve been an ideal test bed for such a model, had anyone thought to track the data. But I think there’s still plenty of room here for a net improvement in quality of conference life at future meetings, with a bit more healthy walking allowing everyone to get fed faster and still make it back for the next session. The APS should consider handing out some grants to study this issue, and maybe a Focus Session at the next March Meeting (in Minneapolis). Just remember to acknowledge this post when you write it up…
This bit of silliness brought to you by the winter storm blanketing the Northeast at the moment, which has actually knocked out power to our neighborhood, but we have a standby generator that’s allowed the restoration of Internet service to Chateau Steelypips. If you want to see whether I get even goofier as the storm continues, here’s a button:
And if you’d like to suggest refinements or extensions of the model, the comments will be open:
A Proposal for Social Physics
This post reminds me of every time I got in the elevator in the CS department and wondered “how many of the people in this elevator are thinking about the algorithm the elevator is running” lolz.
With lunch, there is of course the balance of walking/waiting time (to which I think the best solution is "buy something in the morning on your way to the convention center") but what always struck me (in general, I wasn't at Mar2023) was that the physicist density for dinner is also very strong near the convention center and drops off much more sharply than r^(-2).
And there are so many other things that numerical calculations could model about conferences, starting with assigning sessions to rooms based on projected demand. Or, arranging of chairs in the conference room to minimize the phenomenon of middle-of-the-aisle chairs free up front, two-deep standees at the back. (My suggestion to that is to create more aisles by removing some chairs, but I don't know the optimal chair/aisle balance.) And while I'm complaining about conference inefficiencies, why hasn't any convention center ever figured out how to efficiently sell coffee?