2019/02/13 by Hugo Lewi Hammer, Hammer, Hugo Lewi, Anis Yazidi +3
Mathematics · #FOS: Computer and information sciences #Methodology (stat.ME) #stat.ME
paper · pdf · doi:10.48550/arxiv.1902.05428
arXiv admin note: text overlap with arXiv:1901.04681
arxiv created 2019/02/13 · arxiv updated 2019/02/15
Estimation of quantiles is one of the most fundamental real-time analysis tasks. Most real-time data streams vary dynamically with time and incremental quantile estimators document state-of-the art performance to track quantiles of such data streams. However, most are not able to make joint estimates of multiple quantiles in a consistent manner, and estimates may violate the monotone property of quantiles. In this paper we propose the general concept of *conditional quantiles* that can extend incremental estimators to jointly track multiple quantiles. We apply the concept to propose two new estimators. Extensive experimental results, on both synthetic and real-life data, show that the new estimators clearly outperform legacy state-of-the-art joint quantile tracking algorithm and achieve faster adaptivity in dynamically varying data streams.