Relative Error Streaming Quantiles
Abstract
Approximating ranks, quantiles, and distributions over streaming data is a central task in data analysis and monitoring. Given a stream of $n$ items from a data universe $\mathcal{U}$ equipped with a total order, the task is to compute a sketch (data structure) of size $\mathrm{poly}(\log(n), 1/\varepsilon)$. Given the sketch and a query item $y \in \mathcal{U}$, one should be able to approximate its rank in the stream, i.e., the number of stream elements smaller than $y$. Most works to date focused on additive $\varepsilon n$ error approximation, culminating in the KLL sketch that achieved optimal asymptotic behavior. This paper investigates multiplicative $(1\pm\varepsilon)$error approximations to the rank. The motivation stems from practical demands to understand the tails of distributions, and hence for sketches to be more accurate near extreme values. The most spaceefficient algorithms that can be derived from prior work store either $O(\log(\varepsilon^2 n)/\varepsilon^2)$ or $O(\log^3(\varepsilon n)/\varepsilon)$ universe items. This paper presents a randomized sketch of size $O(\log^{1.5}(\varepsilon n)/\varepsilon)$ that achieves a $1\pm\varepsilon$ multiplicative error guarantee, without prior knowledge of the stream length or dependence on the size of the data universe. This is within a $O(\sqrt{\log(\varepsilon n)})$ factor of optimal. Moreover, our sketch is fully mergeable, which makes it suitable for a parallel or distributed environment.
 Publication:

arXiv eprints
 Pub Date:
 April 2020
 arXiv:
 arXiv:2004.01668
 Bibcode:
 2020arXiv200401668C
 Keywords:

 Computer Science  Data Structures and Algorithms;
 F.2.2
 EPrint:
 Major improvement of the manuscript: better streaming result, added proof of full mergeability and an analysis with extremely small failure probability. 43 pages, 2 figures