Data-Driven Ghosting using Deep Imitation Learning
- Added: 5 months ago
- Current state-of-the-art sports statistics compare players and teams to league average performance. For example, metrics such as “Wins-above-Replacement” (WAR) in baseball, “Expected Point Value” (EPV) in basketball and “Expected Goal Value” (EGV) in soccer and hockey are now commonplace in performance analysis. Such measures allow us to answer the question “how does this player or team compare to the league average?” Even “personalized metrics” which can answer how a “player’s or team’s current performance compares to its expected performance” have been used to better analyze and improve prediction of future outcomes. These measures have enhanced our ability to analyze, compare and value performance in sport. But they are inherently limited because they are tied to a discrete outcome of a specific event. For example, EPV for basketball focuses on estimating the probability of a player making a shot based on the current situation, and is learnt off enormous amounts of historical data. The general use case is then to aggregate these outcomes, and compare and rank them to see how various players and teams compare to each other. In contrast, what we’d really like to know is how teams create time and space for scoring opportunities at the fine-grain level.
Link to publication page: www.disneyresearch.com/publication/data-driven-ghosting/
- Science & Technology
- Runtime: 2:36
- Disney Research Pittsburgh Ghosting innovation research science technology tech Deep Learning Deep Imitation Learning athletics sports ESPN Sloan Sports Analytics Conference