This is a review of the data driven ghosting with deep imitation learning research conducted by Hoang M. Le, Peter Carr, Yisong Yue, and Patrick Lucey.
Statistics typically compare players and teams to the league average. However, this can be very limiting. A coach may want to compare their team against an opposing team rather than the league average. Or they might want to determine what characteristics allows an opponent to dominate in a certain aspect of the game in order to adopt that style for themselves or develop a strategy to defend against it. A data-driven ghosting technique can be useful in these situations.
A network employing deep imitation learning can be used. This type of network allows ghosted players to anticipate the movements of their teammates as well as the moves the opposition will make. Imitation Learning is also known as learning from demonstrations. Machines learn good policies and skills by observing expert behavior. Deep learning involves learning complex layers of information hidden within data. The network learns from player tracking and event data from 100 games played in a professional soccer league. Additional information was inputted into the system including the main role of each player, coordinates of each player and the ball, as well the distance and angle of each player towards the ball and goal.
Analysts can use this network to look at how different teams stack up against each other. It can help determine which team is expected to win, especially in cases where the two teams are evenly matched.
This network can now be used to look at the consequences that would emerge from hypothetical situations. This provides coaches with a safe way to examine different types of plays without putting their players in harm’s way. Coaches can see how their team compares to the average team in order to help determine their specific strengths and weaknesses.
The data driven ghosting scenario can be changed so that instead of reflecting an average team a specific team is reflected instead. Specific aspects of the teams, such as defensive behavior, can be compared. This gives coaches the ability to look at how other teams would handle different situations or how they might react to the plays they plan to use in their next match. From this they formulate a game plan to counter the decisions the other team is expected to make. Comparing characteristics between teams can help decipher what makes their opponent tick allowing coaches to develop effective play strategies.
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.