Diner Dash

Diner Dash是一个用于评估高维动作空间中策略学习算法性能的轻量级基准任务。该数据集由新加坡国立大学创建，旨在解决传统基准如Atari游戏在动作空间维度上的局限性。Diner Dash具有层次任务结构和57维动作空间，模拟了一个餐厅经营环境，其中玩家需控制服务员服务尽可能多的顾客。数据集创建过程中，通过引入分解策略图建模（DPGM）算法，结合图建模和深度学习，实现了领域知识的明确嵌入。该数据集适用于研究复杂任务中的策略学习，特别是在高维动作空间和状态空间中的应用，如交通灯控制等实际问题。

Diner Dash is a lightweight benchmark task for evaluating the performance of policy learning algorithms in high-dimensional action spaces. This dataset was created by the National University of Singapore, aiming to address the limitations of traditional benchmarks such as Atari games in terms of action space dimensionality. Diner Dash features a hierarchical task structure and a 57-dimensional action space, simulating a restaurant management environment where players control waitstaff to serve as many customers as possible. During the dataset creation process, the decomposed policy graph modeling (DPGM) algorithm was introduced, combining graph modeling and deep learning to achieve explicit embedding of domain knowledge. This dataset is suitable for research on policy learning in complex tasks, especially for applications in high-dimensional action and state spaces such as practical problems like traffic light control.