Data underlying the publication: Structure-preserving contrastive learning for spatial time series

This dataset includes the resulting data of the research: Structure-preserving contrastive learning for spatial time series. It includes precomputed distance matrices, logs and results from hyperparameter grid search, trained encoder checkpoints, as well as evaluation metrics for UEA classification and traffic prediction tasks. The research is experimental and focuses on enhancing self-supervised contrastive learning by preserving fine‐grained spatio-temporal similarity structures. The proposed methods are applied to public UEA archive datasets of multivariate time series and specialised macro- and micro-traffic datasets. The scripts that produced these data are open-sourced at https://github.com/Yiru-Jiao/SPCLT

本数据集包含一项研究的成果数据：面向空间时序的保结构对比学习（Structure-preserving contrastive learning for spatial time series）。其涵盖预计算距离矩阵、超参数网格搜索的日志与结果、已训练的编码器检查点，以及用于UEA分类与交通预测任务的评估指标。该研究属于实验性研究，旨在通过保留细粒度时空相似性结构，增强自监督对比学习的效果。所提出的方法已应用于公开的UEA多变量时序档案数据集，以及专用的宏观与微观交通数据集。生成上述数据的代码脚本已在https://github.com/Yiru-Jiao/SPCLT 开源。