Efficient framework for operating on data sketches

Abstract

We study the problem of analyzing massive data streams based on concise data sketches. Recently, a number of papers have investigated how to estimate the results of set-theory operations based on sketches. In this paper we present a framework that allows to estimate the result of any sequence of set-theory operations. The starting point for our solution is the solution from 2021. Compared to this solution, the newly presented sketching algorithm is much more computationally efficient as it requires on average O (log n ) rather than O ( n ) comparisons for n stream elements. We also show that the estimator dedicated to sketches proposed in that reference solution is, in fact, a maximum likelihood estimator.