“…Due to these challenges, most previous work in sports analytics have focused on relatively small datasets [14], [15], [16], did not build predictive models [17], [8], [18], used coarse aggregate statistics that do not model specific in-game scenarios [19], did not focus on adversarial team environments [20], do not model spatial information [21], [22], or require extensive manual annotation [5]. In contrast, we are interested in making in-game predictions of near-term events over a large selection of adversarial in-game scenarios.…”