Streaming Algorithms for Partitioning Integer Sequences

Abstract: We study the problem of partitioning integer sequences in the one-pass data streaming model. Given is an input stream of integers $X \in {0, 1, \dots, m }^n$ of length $n$ with maximum element $m$, and a parameter $p$. The goal is to output the positions of separators splitting the input stream into $p$ contiguous blocks such that the maximal weight of a block is minimized. We show that computing an optimal solution requires linear space, and we design space efficient $(1+\epsilon)$-approximation algorithms for this problem following the parametric search framework. We demonstrate that parametric search can be successfully applied in the streaming model, and we present more space efficient refinements of the basic method. All discussed algorithms require space $O( \frac{1}{\epsilon} \mathrm{polylog} (m,n,\frac{1}{\epsilon}))$, and we prove that the linear dependency on $\frac{1}{\epsilon}$ is necessary for any possibly randomized one-pass streaming algorithm that computes a $(1+\epsilon)$-approximation.