MementoML

MementoML数据集由华沙理工大学和华沙大学创建，旨在通过评估7种流行机器学习算法在39个OpenML数据集上的性能，研究超参数对算法性能的影响。数据集涵盖了从500到100000个观测值，特征数不超过5000，主要用于二分类问题。创建过程中，每个数据集被分为20个训练/测试Bootstrap对，每个模型在特定的超参数集上训练20次。该数据集适用于机器学习的元学习研究，旨在发现最佳超参数默认值和合理超参数空间，以及特定超参数的重要性。

The MementoML dataset was developed by Warsaw University of Technology and University of Warsaw to study the impact of hyperparameters on machine learning algorithm performance. It evaluates 7 popular machine learning models across 39 OpenML datasets, each of which contains 500 to 100,000 observations and no more than 5,000 features, and is primarily intended for binary classification tasks. During the construction of MementoML, each source dataset was split into 20 training/test bootstrap pairs, and each of the 7 models was trained 20 times using a specific hyperparameter configuration. This dataset is suitable for machine learning meta-learning research, with the objective of discovering optimal default hyperparameters, reasonable hyperparameter search spaces, and the importance of individual hyperparameters.