Bottom-k and Priority Sampling, Set Similarity and Subset Sums with Minimal Independence

TL;DR

This paper demonstrates that bottom-k sampling can reliably estimate set similarities and subset sums with only 2-independent hash functions, achieving near-optimal error bounds, and extends the approach to weighted sets using priority sampling.

Contribution

It proves that bottom-k sampling maintains accurate estimations with minimal hash independence, reducing the required independence from 8 to 2, and extends the analysis to weighted sets with priority sampling.

Findings

Expected relative error is O(1/√(fk)) with 2-independence.

Bottom-k sampling achieves near-constant error compared to fully random hashing.

Priority sampling effectively handles weighted sets with strong concentration bounds.

Abstract

We consider bottom-k sampling for a set X, picking a sample S_k(X) consisting of the k elements that are smallest according to a given hash function h. With this sample we can estimate the relative size f=|Y|/|X| of any subset Y as |S_k(X) intersect Y|/k. A standard application is the estimation of the Jaccard similarity f=|A intersect B|/|A union B| between sets A and B. Given the bottom-k samples from A and B, we construct the bottom-k sample of their union as S_k(A union B)=S_k(S_k(A) union S_k(B)), and then the similarity is estimated as |S_k(A union B) intersect S_k(A) intersect S_k(B)|/k. We show here that even if the hash function is only 2-independent, the expected relative error is O(1/sqrt(fk)). For fk=Omega(1) this is within a constant factor of the expected relative error with truly random hashing. For comparison, consider the classic approach of kxmin-wise where we use k hash independent functions h_1,...,h_k, storing the smallest element with each hash function. For kxmin-wise there is an at least constant bias with constant independence, and it is not reduced with larger k. Recently Feigenblat et al. showed that bottom-k circumvents the bias if the hash function is 8-independent and k is sufficiently large. We get down to 2-independence for any k. Our result is based on a simply union bound, transferring generic concentration bounds for the hashing scheme to the bottom-k sample, e.g., getting stronger probability error bounds with higher independence. For weighted sets, we consider priority sampling which adapts efficiently to the concrete input weights, e.g., benefiting strongly from heavy-tailed input. This time, the analysis is much more involved, but again we show that generic concentration bounds can be applied.