Synthetic Historical Dataset

该数据集包含了第一年收集的日志数据，用于估算第二年目标时间点新政策的未来价值。数据集融合了季节性和非平稳特性，通过抽样时间戳和上下文向量，合成预期的奖励函数，并应用记录策略生成。此外，该数据集还包括季节性效应，旨在模拟非平稳性。数据规模根据样本数量而定（对于未来离策略评估（F-OPE）样本量为1000，对于未来离策略学习（F-OPL）样本量为8000）。该数据集的任务包括未来离策略评估（F-OPE）和未来离策略学习（F-OPL）。

This dataset contains logged data collected in the first year, which is employed to estimate the future value of a new policy at the target time point in the second year. It integrates both seasonal and non-stationary properties, and is generated by first sampling timestamps and context vectors to synthesize the expected reward function, then applying the logging policy. Additionally, the dataset incorporates seasonal effects to simulate non-stationarity. The scale of the dataset varies based on the sample size: 1000 samples for Future Off-Policy Evaluation (F-OPE) and 8000 samples for Future Off-Policy Learning (F-OPL). The tasks supported by this dataset include Future Off-Policy Evaluation (F-OPE) and Future Off-Policy Learning (F-OPL).