OR-Gym: A Reinforcement Learning Library forOperations Research Problems
基于强化学习的优化问题探索
Abstract
Reinforcement learning (RL) has been widely applied to game-playing and surpassedthe best human-level performance in many domains, yet there are few use-cases inindustrial or commercial settings. We introduce OR-Gym, an open-source library fordeveloping reinforcement learning algorithms to address operations research problems.In this paper, we apply reinforcement learning to the knapsack, multi-dimensional binpacking, multi-echelon supply chain, and multi-period asset allocation problems, andbenchmark the RL solutions against MILP and heuristic models. These problems areused in logistics, finance, engineering, and are common in many business operationsettings. We develop environments based on prototypical models in the literature andimplement various optimization and heuristic models in order to benchmark the RLresults. By re-framing a series of classic optimization problems as RL tasks, we seek toprovide a new tool for the operations research community, while also opening those inthe RL community to many of the problems and challenges in the OR field.Keywords: Machine Learning, Reinforcement Learning, Optimization, Operations Research, Robust Optimization