anonadata/CUVET-policy

CUVET-Policy数据集是一个由在线广告平台通过两周的A/B测试收集的匿名化数据集，包含了五种不同处理的随机分配参数。该数据集旨在学习一个政策，将连续的治疗策略分配给用户，以期在考虑成本约束的同时，生成比参考治疗更多的预期价值。数据集规模为86.7M行，每行代表一个用户，包含连续的用户特征、标签（价值和成本）以及治疗变量。数据集被随机划分为训练集和测试集，训练集还包含一个额外的二元列，用于分离论文实验中使用的验证集。数据集遵循CC-BY-NC-SA 4.0许可证发布，可用于决策约束下的学习模型、不确定性下的优化方法和利用噪声实验数据改进预测算法等。

The CUVET-Policy dataset is an anonymized dataset collected by an online advertising platform through a two-week A/B test with five different randomly assigned treatment parameters. The dataset is designed to learn a policy that assigns a continuous treatment strategy to users, aiming to generate more expected value than the reference treatment while considering cost constraints. The dataset is 86.7M rows, with each row representing a user, including continuous user features, labels (value and cost), and a treatment variable. The dataset is randomly split into training and test sets, with the training set including an additional binary column to separate the validation set used in the papers experiments. The dataset is released under the CC-BY-NC-SA 4.0 license and can be used for learning models under decision constraints, variance-aware methods, and causal methods using noisy experimental data.