xenosaac/alphahack-devpost
收藏资源简介:
AlphaHack Devpost数据集是一个用于预测黑客马拉松结果的因子工程数据集。它通过爬取2019年至2026年间Devpost的公开项目和活动页面构建而成。数据集包含265,088个项目行、11,972个活动、1,409个带有评判标准的活动、28,708行Sonnet注释的评分标准、500对A/B比较、100个精选高质量获奖示例以及一个包含101,682行×151列的工程特征parquet文件。数据集旨在用于研究黑客马拉松生态系统中的竞争动态、公共竞赛的因子建模、AlphaHack模型的可重复性以及LLM生成评分标准的分布外分析。数据集的使用受到一些限制,如仅包含英语内容、企业赞助活动过度代表等。
The AlphaHack Devpost Dataset is a factor-engineering dataset for hackathon-outcome prediction, built by crawling Devposts public project and event pages between 2019 and 2026. It includes 265,088 project rows across 10 shards, 11,972 events, 1,409 events with judging criteria, 28,708 Sonnet-annotated rubric scores, 500 pairwise A/B comparisons, 100 curated top-quality winner exemplars, and an engineered feature parquet with 101,682 rows × 151 columns. The dataset is intended for research on competition dynamics in hackathon ecosystems, factor modeling for public competitions, reproducibility of the AlphaHack models, and out-of-distribution analysis of LLM-generated rubric scores. Limitations include English-language content only, over-representation of corporate-sponsored events, and other documented biases.





