Very interesting! With some manual or automatic annotation, this is one step closer to spatiotemporal player data, which opens up whole new boundaries of analys

Very interesting! With some manual or automatic annotation, this is one step closer to spatiotemporal player data, which opens up whole new boundaries of analysis. As a point of comparison, there's a private NBA dataset (based on imaging and manual annotation) which shows the locations of shots made, annotated with the player name and whether they failed or succeeded - from this, you can do really cool things like Miller et. al. 2014 [1], who use dimensionality reduction to find patterns in players' preferred shot locations, potentially allowing you to create teams whose offensive abilities are complementary from a spatial perspective. [1] http://jmlr.org/proceedings/papers/v32/miller14.pdf http://jmlr.org/proceedings/papers/v32/miller14.pdf