来自$\sim$ 1400万条VVV光曲线的RRLyrics自动编目：传统机器学习能走多远？
Automatic Catalog of RRLyrae from $\sim$ 14 million VVV Light Curves: How far can we go with traditional machine-learning?
摘要：
使用RRLyrics（RRL）创建隆起的3D地图是VVV（X）调查的主要目标之一。绝大多数正在分析的来源要求使用自动程序。在这种情况下，以前的作品介绍了使用机器学习（ML）方法的变星分类。我们的目标是开发和分析一个自动程序，基于ML，识别RRL的VVV调查。此过程将用于生成整合在调查中的多个瓦片上的可靠目录。在光变曲线的重建之后，我们提取了一组基于周期和强度的特征。我们第一次使用一个新的子集的伪彩色功能。我们讨论了定义自动管道所需的所有适当步骤：质量测量的选择;采样程序;分类器设置和模型选择。作为最终的结果，我们构建了一个集成分类器，平均召回率为0.48，平均精度为0.86超过15个瓷砖。我们还提供了我们处理过的数据集和候选RRL的目录。也许最有趣的是，从基于光度宽带数据的分类角度来看，我们的研究结果表明，颜色是RRL的一种信息特征类型，应该考虑通过ML进行自动分类。我们还认为，召回率和精度在这两个表和曲线是高质量的指标，这个高度不平衡的问题。此外，我们表明，我们的VVV数据集，有良好的估计，重要的是要使用原始分布比减少样本与人工平衡。最后，我们表明，集成分类器的使用有助于解决关键的模型选择步骤，并在RRL的识别中的大多数错误与低质量的一些来源的观测或难以解决的RRL-C型给定的日期。
英文摘要：
The creation of a 3D map of the bulge using RRLyrae (RRL) is one of the main goals of the VVV(X) surveys. The overwhelming number of sources under analysis request the use of automatic procedures. In this context, previous works introduced the use of Machine Learning (ML) methods for the variable star classification. Our goal is the development and analysis of an automatic procedure, based on ML, for the identification of RRLs in the VVV Survey. This procedure will be use to generate reliable catalogs integrated over several tiles in the survey. After the reconstruction of light-curves, we extract a set of period and intensity-based features. We use for the first time a new subset of pseudo color features. We discuss all the appropriate steps needed to define our automatic pipeline: selection of quality measures; sampling procedures; classifier setup and model selection. As final result, we construct an ensemble classifier with an average Recall of 0.48 and average Precision of 0.86 over 15 tiles. We also make available our processed datasets and a catalog of candidate RRLs. Perhaps most interestingly, from a classification perspective based on photometric broad-band data, is that our results indicate that Color is an informative feature type of the RRL that should be considered for automatic classification methods via ML. We also argue that Recall and Precision in both tables and curves are high quality metrics for this highly imbalanced problem. Furthermore, we show for our VVV data-set that to have good estimates it is important to use the original distribution more than reduced samples with an artificial balance. Finally, we show that the use of ensemble classifiers helps resolve the crucial model selection step, and that most errors in the identification of RRLs are related to low quality observations of some sources or to the difficulty to resolve the RRL-C type given the date.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
