HPO-B
收藏资源简介:
HPO-B是一个大规模可重现的基准数据集,用于黑盒超参数优化(HPO),由弗莱堡大学等机构创建。该数据集包含从OpenML存储库中收集的176个搜索空间和196个数据集,总计约640万个超参数评估。HPO-B旨在解决HPO领域中由于计算资源需求巨大而未解决的核心问题,通过提供一个公平的比较平台,促进HPO算法的快速微调和转移学习。数据集的创建过程包括从OpenML下载实验数据,进行数据清洗、预处理和组织,确保数据集的可重现性和准确性。HPO-B的应用领域主要集中在机器学习社区,特别是在超参数优化和自动机器学习(AutoML)领域,旨在通过提供一个标准化的基准,加速HPO方法的研究和开发。
HPO-B is a large-scale reproducible benchmark dataset for black-box hyperparameter optimization (HPO), developed by institutions including the University of Freiburg. This dataset encompasses 176 search spaces and 196 datasets collected from the OpenML repository, totaling approximately 6.4 million hyperparameter evaluations. HPO-B aims to address the core unresolved challenges in the HPO domain stemming from the substantial computational resource requirements, by offering a fair comparison platform to promote rapid fine-tuning and transfer learning of HPO algorithms. The dataset creation process includes downloading experimental data from OpenML, followed by data cleaning, preprocessing and organization to ensure the reproducibility and accuracy of the dataset. The application scenarios of HPO-B are mainly focused on the machine learning community, especially in the fields of hyperparameter optimization and automated machine learning (AutoML), with the goal of accelerating the research and development of HPO methods by providing a standardized benchmark.

- 1HPO-B: A Large-Scale Reproducible Benchmark for Black-Box HPO based on OpenML弗莱堡大学 · 2021年



