Data-streams and histograms

Abstract

Histograms have been used widely to capture data distribution, to represent the data by a small number of step functions. Dynamic programming algorithms which provide optimal construction of these histograms exist, albeit running in quadratic time and linear space. In this paper we provide linear time construction of 1 + ε approximation of optimal histograms, running in polylogarithmic space.