Numerical Evaluation of Baseball Players on Parallel Computational Environments

2014 Fiscal Year Final Research Report

• Project/Area Number: 24650404
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Sports science
• Research Institution: National Agency for the Advancement of Sports and Health
• Principal Investigator: OSAWA Kiyoshi 独立行政法人日本スポーツ振興センター国立スポーツ科学センター, スポーツ科学研究部, 契約研究員 (60601135)
• Co-Investigator(Renkei-kenkyūsha): AIDA Kento 国立情報学研究所, アーキテクチャ科学研究系, 教授 (80247212)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: ハイパフォーマンスコンピューティング / オペレーションズリサーチ

Outline of Final Research Achievements

A probability distribution of runs scored in a baseball game and a winning percentage of the game are able to be computed by multiplying matrices, which are defined by some results of players.
We proposed the method to evaluate players by the relation between the winning percentage and their results. The existing model to compute the probability distributions was based on only batting results, so that fielding results were not applied to the computation. We incorporated fielding results into the extended model.
The numerical experiment indicated that impact of difference in winning percentages, which were calculated from the model with fielding results of players who have good and poor fielding skill, corresponded to about 2 wins for a season. While this problem had large computational complexity due to adding fielding results to the model, the parallel computational environment enabled us to find an optimal batting order among about 210 million orders in 1 hour 22 minutes.

Free Research Field

情報工学