An artificial intelligence can quickly assess basketball game data and extract information about the habits, strengths and weaknesses of players, which could prove valuable for coaches, particularly from smaller teams.
at the University of Massachusetts, Amherst, and his colleagues have used data compiled and publicly released by US company to train an AI model. The data included the 3D location of players and the ball throughout games in the 2015/16 National Basketball Association (NBA) season, and the training took about 3 hours.
The researchers then gave their AI a test data set about a fifth of the size of the training set. It contained information about where basketball players spent most time on the court, where they took shots from, and how this behaviour varied depending on who was near them and at what point in the game it was. The AI took just a few seconds to generate insights that Rodr铆guez says coaches could use to gain a competitive advantage.
Advertisement
For example, if a certain player tends to score well when shooting from one side of the court, but poorly from the other, then a coach could direct players to block them only when they posed a risk, says Rodr铆guez. Or if a player tends to take long-distance shots at the end of the game, a coach could get more players to focus on them in the final quarter.
The researchers have uploaded their refined AI model to so that anyone can use it to analyse similar spatial basketball data. Rodr铆guez says it would also be possible for others to create AI models to create that spatial data from raw video, just as Second Spectrum does.
Second Spectrum hasn鈥檛 responded to a request for comment.
at Davidson College in North Carolina says collection and analysis of statistics is common at the top level in most professional sports. 鈥淚f I can find something that you haven鈥檛 found, then that can be a competitive edge,鈥� says Chartier, who has worked as a consultant for numerous professional sports teams.
鈥淚f you鈥檙e a massive underdog to another team, it鈥檚 unlikely that analysing data is gonna suddenly transport you into being able to win the game,鈥� he says. 鈥淏ut it can make a real difference, and it can really make a difference in tight games; that鈥檚 one of the biggest places that machine learning is employed.鈥�
Rodr铆guez says teams with limited computing power and technical expertise will now be able to access the same resources thanks to open-source data libraries that 鈥渄o all the dirty work for you鈥�.
鈥淵ou can just literally just download a library and a model that someone else has made, and there you go, it鈥檚 done. It鈥檚 very accessible,鈥� says Rodr铆guez, who did the work while at the University of California, San Diego.
This newfound accessibility means high-tech analysis will inevitably spread to lower levels and even amateur leagues, says Chartier.
This may lead to an overreliance on data, says Rodr铆guez, who gives the example of the Houston Rockets basketball team in one particular season.
鈥淭he way they played was purely analytics driven,鈥� he says. 鈥淚t was just 鈥榳e鈥檙e giving the ball to our best player and we鈥檙e putting everyone else on the three-point line, and we鈥檙e going to run the play that gives us the most expected points over and over again鈥�. It was obviously working, but it was maybe the most boring thing I鈥檝e ever watched.鈥�
AIs like the one in this study can also be used in football. Second Spectrum, for example, has contracts with the NBA and the English football Premier League to use AI to collect data. But it is easier to apply the information gleaned to sports in which the games are divided up into a greater number of discrete periods of play and have higher scoring, like American football and baseball, says Rodr铆guez.
arXiv
