StarEmbed/ZTF_40k
收藏资源简介:
StarEmbed基准数据集是一个天文时间序列数据集,包含约40,000个多波段ZTF(Zwicky Transient Facility)光变曲线,用于变星观测研究。数据集中每个变星记录包括:sourceid(源标识符)、bands_data(包含g、i、r波段的字典结构,每个波段存储光变曲线的目标亮度、误差、时间间隔、观测时间和观测数量)、period(变星周期,以天为单位)、class_str(变星类别标签,如EW、EA、RRab等)、ra(赤经)和dec(赤纬)。数据集分为训练集、验证集、测试集和异常集(anom),其中异常集包含分布外类别,用于异常检测基准测试。数据基于ZTF观测和Catalina Surveys Periodic Variable Star Catalog整理,适用于时间序列基础模型和天文AI研究。
The StarEmbed benchmark dataset is an astronomical time series dataset containing approximately 40,000 multi-band ZTF (Zwicky Transient Facility) light curves for variable star observation research. Each variable star record includes: sourceid (source identifier), bands_data (a dictionary structure with g, i, r bands, each storing the light curves target magnitude, error, time gaps, observation times, and observation count), period (variable star period in days), class_str (variable star class label, such as EW, EA, RRab, etc.), ra (right ascension), and dec (declination). The dataset is split into train, validation, test, and anom sets, with the anom set containing out-of-distribution classes for anomaly detection benchmarking. It is curated from ZTF observations and the Catalina Surveys Periodic Variable Star Catalog, suitable for time series foundation models and astronomical AI research.




