Computing an Optimal Pitching Strategy in a Baseball At-Bat
Computing an Optimal Pitching Strategy in a Baseball At-Bat
复制标题
DOI：
10.32473/flairs.36.133346
复制
发表时间：
2021-10
期刊：
影响因子：
--
通讯作者：
Connor Douglas;Everett Witt;Mia Bendy;Yevgeniy Vorobeychik
中科院分区：
文献类型：
--
作者：
Connor Douglas;Everett Witt;Mia Bendy;Yevgeniy Vorobeychik
在过去的十年里，定量分析领域已经改变了体育界。到目前为止，这些分析方法的核心是统计，描述什么是什么，过去是什么，同时利用这些信息来推动关于未来做什么的决策。然而，由于我们经常将足球、曲棍球和棒球等团队运动视为双赢的对决，将它们建模为零和游戏似乎是很自然的。我们提出了这样一个棒球击球模型，它是投手和击球手之间的对决。具体地说，我们提出了一种新的模型来描述这种相遇，这是一个零和随机博弈，其中击球手的目标是上垒，而投手想要防止的结果。这场比赛的价值是上垒百分比(即击球手上垒的概率)。原则上，这个随机博弈可以用经典的方法来解决。主要的技术挑战在于预测作为投手意图的函数的投球位置的分布，预测如果击球手决定在投球时挥杆的结果的分布，以及表征特定击球手的耐心水平。我们通过提出新的投手和击球手的表示以及用于结果预测的新的深层神经网络结构来解决这些挑战。我们使用2015到2018年美国职业棒球大联盟赛季的Kaggle数据进行的实验证明了所提出的方法的有效性。
The field of quantitative analytics has transformed the worldof sports over the last decade. To date, these analytic ap-proaches are statistical at their core, characterizing what isand what was, while using this information to drive decisionsabout what to do in the future. However, as we often viewteam sports, such as soccer, hockey, and baseball, as pairwisewin-lose encounters, it seems natural to model these as zero-sum games. We propose such a model for a baseball at-bat,which is a matchup between a pitcher and a batter. Specifi-cally, we propose a novel model of this encounter as a zero-sum stochastic game, in which the goal of the batter is to geton base, an outcome the pitcher aims to prevent. The valueof this game is the on-base percentage (i.e., the probabilitythat the batter gets on base). In principle, this stochastic gamecan be solved using classical approaches. The main techni-cal challenges lie in predicting the distribution of pitch loca-tions as a function of pitcher intention, predicting the distri-bution of outcomes if the batter decides to swing at a pitch,and characterizing the level of patience of a particular batter.We address these challenges by proposing novel pitcher andbatter representations as well as a novel deep neural networkarchitecture for outcome prediction. Our experiments usingKaggle data from the 2015 to 2018 Major League Baseballseasons demonstrate the efficacy of the proposed approach.