eldieguinpo/criteo-1M

Criteo 1M 是一个从完整的Criteo Display Advertising Challenge数据集中均匀随机抽取的1,000,000行样本。采样过程保留了原始点击/未点击的基础率，样本点击率为25.12%，与完整数据集约25.6%的点击率相近，因此该样本适合作为内存友好型替代品，用于反映原始点击率任务的自然类别平衡。数据集包含40列：目标变量label（二进制，1表示广告被点击，0表示未点击）、13个数值特征（I1到I13，存储为double类型，包含缺失值，原始值为非负整数，但语义未公开）、26个分类特征（C1到C26，已整数编码为int64类型，代表类别ID，不应视为数值，适合使用嵌入、独热或目标编码处理）。数据集适用于表格分类任务，特别是点击率预测，规模为1M到10M之间，许可证为CC-BY-NC-SA 4.0。

Criteo 1M is a uniform random sample of 1,000,000 rows drawn from the full Criteo Display Advertising Challenge dataset. The sampling preserves the original click / no-click base rate, with a click rate of 25.12% in this sample, closely matching the full datasets rate of approximately 25.6%. This makes the file suitable as a drop-in stand-in for the full dataset when a memory-friendly version is needed that still reflects the natural class balance of the original CTR task. The dataset consists of 40 columns: a binary target label (1 = ad clicked, 0 = not clicked), 13 integer/numerical features (I1 to I13, stored as double with missing values, originally non-negative integers with unpublished semantics), and 26 categorical features (C1 to C26, integer-encoded as int64 representing category IDs, not to be treated as numeric, suitable for embedding, one-hot, or target encoding). It is designed for tabular-classification tasks, particularly click-through rate prediction, with a size category of 1M<n<10M and licensed under CC-BY-NC-SA 4.0.