Introducing Deep Reinforcement Learning Framework for Intelligent Baseball Scheduling
- Program: Computer Science
- Course: Other
Project Description:
Scheduling sports matches is still a very challenging problem. The goal
of sports scheduling is not only to create fair and balanced baseball
schedules but also to satisfy various kaleidoscopic constraints. While
advancements have been made through optimization approaches such as
semi-definite programming, hybrid approaches, and heuristic methods,
these approaches are limited to fixed scenarios and do not completely
consider real-time scheduling demands. Motivated by these limitations,
our work consists of the following:
1) Developed a custom constraint library for the required and desired
constraints
2) Proposed a reinforcement learning framework between the baseball
league schedule and the RL agents to create satisfying baseball
scheduling schedules, which is based off a Markov Decision Process.
Our simulative experiments with specific constraints of the 2022 High A
Central season suggests that the RL agents were able to achieved
consistently high schedule satisfaction score (over 70%) on average after
training with Deep Q Networks.
