Basketball Pass & Shot Prediction
(Instructions at bottom)
Instructions:
  • This demo predicts shot and pass probabilities for the blue team.
  • The current ball handler is shaded dark blue.
  • The other players are shaded light blue.
  • The other team is shaded red.
  • The thickness of light blue lines indicate the magnitude of the pass probability from the ball handler to each player.
  • The thickness of the grey line indicates the magnitude of the shot probability of the ball handler.

  • You can drag any player.
  • You can right click on any blue player to set it as the ball handler.
  • You can click on the "Scenario" buttons set the field to different game states.

  • One feet is roughly 13 pixels.



    • About Our Model:
  • Our approach is described in detail in [1].
  • Our model is a variant of a Conditional Random Field [3].
  • Conditioned on the game state, our model predicts a probability distribution for the ball handler passing to each teammate or not passing at all.
  • The current version of our model accounts for factors such as:
    • How far is the teammate from the ball handler?
    • How close are defenders to the passing lane?
    • How much closer is the teammate to the opponent's basket?
  • Our model also learns spatial patterns of basketball gameplay, similar to the approach used in [2]
  • Our model is trained using NBA spatiotemporal player tracking data provided by STATS SportsVU.



    • Sports Analytics @ Disney Research Pittsburgh:
  • Yisong Yue
  • Patrick Lucey
  • Peter Carr
  • Alina Bialkowski
  • Iain Matthews



    • References
    [1] Yisong Yue, Patrick Lucey, Peter Carr, Alina Bialkowski, Iain Matthews, "Learning Fine-Grained Spatial Models for Dynamic Sports Play Prediction", IEEE International Conference on Data Mining (ICDM), 2014. [pdf]

    [2] Andrew Miller, Luke Bornn, Ryan Adams, Kirk Goldsberry, "Factorized Point Process Intensities: A Spatial Analysis of Professional Basketball", International Conference on Machine Learning (ICML), 2014. [pdf]

    [3] John Lafferty, Andrew McCallum, Fernando Pereira, "Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data", International Conference on Machine Learning (ICML), 2001. [pdf]