Open RL Benchmark
收藏资源简介:
Open RL Benchmark是一个社区驱动的强化学习实验数据集,包含超过25,000次跟踪运行,累计时长超过8年。该数据集不仅包含常规数据如情节回报,还包括所有算法特定的系统指标。数据集涵盖了广泛的RL库和参考实现,特别注重确保每个实验的精确可重现性,提供完整的参数和生成它所使用的依赖版本。此外,Open RL Benchmark还配备了一个命令行界面(CLI),方便获取和生成结果展示图表。该数据集旨在提高研究领域的工作效率和精确性,是首个此类RL基准。
Open RL Benchmark is a community-driven reinforcement learning experimental dataset that encompasses over 25,000 tracked runs with a cumulative duration of more than 8 years. This dataset not only includes conventional data such as episode returns, but also all algorithm-specific system metrics. It covers a broad spectrum of RL libraries and reference implementations, with a particular emphasis on guaranteeing the exact reproducibility of each experiment by providing complete experimental parameters and the versions of dependencies used to conduct those experiments. Furthermore, Open RL Benchmark is equipped with a command-line interface (CLI) to facilitate data retrieval and the generation of result visualization charts. This dataset aims to improve the efficiency and accuracy of research work in the field, and it stands as the first RL benchmark of its kind.




