New Results for Adaptive and Approximate Counting of Inversions

TL;DR

This paper introduces new adaptive I/O-efficient algorithms for counting inversions in large datasets and provides a linear-time estimation method for the inversion count in general real-valued lists.

Contribution

It presents the first adaptive I/O-model algorithm for inversion counting and extends linear-time estimation techniques to real-valued data.

Findings

I/O complexity is optimized for cases where $K^*=O(NM)$

A linear-time estimator for $K^*$ in real-valued lists

Extension of estimation algorithms from permutations to general real data

Abstract

Counting inversions is a classic and important problem in databases. The number of inversions, $K^*$, in a list $L=(L(1),L(2),\ldots,L(n))$ is defined as the number of pairs $i < j$ with $L(i) > L(j)$. In this paper, new results for this problem are presented: (1) In the I/O-model, an adaptive algorithm is presented for calculating $K^{*}$. The algorithm performs $O(\frac{N}{B}+ \frac{N}{B}\log_{M/B}(\frac{K^*}{NB}))$ I/Os. When $K^{*}=O(NM)$, then the algorithm takes only $O(\frac{N}{B})$ I/Os. This algorithm can be modified to match the state of the art for the comparison based model and the RAM model. (2) In the RAM model, a linear-time algorithm is presented to obtain a tight estimate of $K^*$; specifically, a value which lies with high probability in the range $[(1-\frac{\log N}{N^{1/4}})K^*,(1+\frac{\log N}{N^{1/4}})K^*]$. The state of the art linear-time algorithm works for the special case where $L$ is a permutation, i.e., each $L(i)$ is a distinct integer in the range $[1,N]$. In this paper, we handle a general case where each $L(i)$ is a real number.