Randomly Generated MDPs

该数据集包含了100个随机生成的马尔可夫决策过程（MDPs），每个MDP具有10个状态和5个动作，且对于所有的状态-动作对（i,a），都满足概率P(i|i,a)大于0的条件，同时奖励是有限的。此外，该数据集被用于比较GSQL1和GSQL2算法与Q学习、快速Q学习和双重Q学习算法的性能。数据集的规模覆盖了具有不同状态空间基数（10、50、100、500、1000）的100个MDPs。该数据集的任务是对强化学习算法进行评估。

This dataset contains 100 randomly generated Markov Decision Processes (MDPs). Each MDP has 10 states and 5 actions, and for all state-action pairs (i,a), the transition probability P(i|i,a) is greater than 0, with bounded rewards. Additionally, this dataset is utilized to compare the performance of GSQL1, GSQL2 algorithms against Q-learning, Fast Q-learning, and Double Q-learning algorithms. The dataset covers 100 MDPs with different state space cardinalities: 10, 50, 100, 500, and 1000. The task of this dataset is to evaluate reinforcement learning algorithms.