anon2435/CoPeP-sample

CoPeP Reviewer Sample 是一个紧凑的确定性样本，用于审查员检查文件格式、模式及示例记录，而无需下载完整的 CoPeP 数据集。它包含训练样本和验证集，训练样本保留了主数据集的训练行模式，如名称、序列、序列哈希、任务列（task_0 至 task_9）和行索引。采样方法基于首次观察到的时序任务进行分组，并使用确定性评分（基于序列哈希、行索引和固定种子）选择每组的行（最多10,000行）。此样本不能完全替代完整数据集，因为它不包含任务分割文件。CoPeP Reviewer Sample is a compact deterministic sample for reviewer inspection to check file format, schema, and example records without downloading the full CoPeP dataset. It includes a training sample and a validation set, with the training sample preserving the main datasets train-row schema, such as name, sequence, seq_hash, task columns (task_0 through task_9), and row_idx. Sampling is based on grouping rows by the first observed temporal task, using a deterministic score derived from seq_hash, row_idx, and a fixed seed to select up to 10,000 rows per group. This sample is not a drop-in replacement for the full dataset, as it does not include task split files.