This is a review of the deep reinforcement learning research performed by Jiaxuan Wang, Ian Fox, Jonathan Skaza, Nick Linck, Satinder Singh and Jenna Wiens.
Double teaming (two defensive players focused on one offensive player) is a strategy regularly employed in the NBA. It can result in slowing down strong offensive players; however, it can also be risky since it leaves another player open. Balancing the reward/risk is a very challenging proposition in the NBA and deciding which way to go depends on many different factors. Effective double teaming involves taking into account where and who all the players are, as well as anticipating the ball handler’s next move. The offensive strategy of the opposing team must also be taken into account. A framework using the deep reinforcement learning approach has been developed to determine when it is the right time to double team another player.
This method for learning how to effectively double team uses the reinforcement learning framework. In this framework the player is interacting in their environment to receive an award, such as blocking a shot. Over 643,000 possessions from the previous three seasons were entered into the framework. Each possession began once all players crossed half court and concluded when the shot clock reset. Included are possible factors affecting the decision to double team such as player heights, weights, shooting abilities, current state of the game, shot clock and game clock. This particular framework focuses strictly on whether or not to double team the ball handler.
After running the data through the framework several observations were made. It was determined that double teaming results in a significantly lower field goal percentage for the offense overall. However, it is also more likely that the possession will end in a foul. It was also determined that when a ball handler is double teamed they typically tend to pass or dribble the ball. The observations indicate that it is more beneficial for the player to pass the ball rather than keep the ball themselves.
In the final analysis it was determined that it is important for the double teamer to be positioned in such a way that he blocks a potential pass to the open man. It also indicates that it is better to double team role players rather than star players. This goes against the typical strategy currently employed in the NBA where star players are double teamed more regularly than role players.
This approach gives analysts a more comprehensive framework to employ while working to understand and evaluate defensive plays and which offensive players perform best when double teamed.
Coaches face many tough decisions every game, just one of them being when to double team and when not to. They can look at this research to help clarify or improve their decision making skills in this area. This framework can also be modified to answer other questions regarding offensive and defensive strategy.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Learn about the Groundbreaking Sports Analytics Model coaches and sports analysts are talking about!
Learn all about Sports Analytics here.