mouse-example-dataset
收藏资源简介:
本数据集包含来自强化学习环境FrozenLake的离线轨迹数据,旨在为离线强化学习、策略评估或模型分析提供基准数据。数据集包含四个独立配置(proc_frozenlake_0至proc_frozenlake_3),每个配置具有相同的结构。每个样本包含标准强化学习轨迹字段:动作(action)、时间步(time)、奖励(reward)、回合终止标志(done)、回合索引(episode_index)、任务索引(task_index)和观察状态(observation)。此外,还提供了额外信息字段:转移概率(info_prob)、最优动作值函数估计(info_q_star)和环境真实最优动作值函数(info_env_q_star)。每个配置包含1500个训练样本,数据以结构化表格形式存储,适用于强化学习算法训练、策略分析、值函数研究等任务。
This dataset contains offline trajectory data from the FrozenLake reinforcement learning (RL) environment, and is designed to provide benchmark data for offline reinforcement learning, policy evaluation, or model analysis tasks. The dataset includes four independent configurations (proc_frozenlake_0 to proc_frozenlake_3), each sharing an identical structure. Each sample contains standard RL trajectory fields: action, time step, reward, done flag, episode_index, task_index, and observation. Additionally, it provides extra information fields: transition probability (info_prob), optimal action-value function estimate (info_q_star), and ground-truth optimal action-value function of the environment (info_env_q_star). Each configuration contains 1500 training samples, and the data is stored in structured table format, suitable for tasks such as reinforcement learning algorithm training, policy analysis, and value function research.
数据集概述
数据集名称: mouse-example-dataset
数据集地址: https://huggingface.co/datasets/micahr234/mouse-example-dataset
配置信息:
- 配置名称:
proc_frozenlake_21 - 数据集大小: 198000 字节
- 下载大小: 107146 字节
数据集特征:
action: int64 类型time: int64 类型reward: float32 类型done: int64 类型episode_index: int64 类型task_index: int64 类型observation: int64 类型info_prob: float64 类型info_q_star: float64 列表info_env_q_star: float64 列表
数据集划分:
- 训练集 (train): 1500 个样本,占用 198000 字节
- 数据文件路径:
data/proc_frozenlake_21/train-*
- 数据文件路径:




