[Paper Review] Structure-Aware Sampling: Flexible and Accurate Summarization

Edith Cohen, Graham Cormode+1 more|arXiv (Cornell University)|Feb 25, 2011

Data Management and Algorithms23 references4 citations

TL;DR

This paper proposes structure-aware sampling, a variance-optimal sampling method that improves accuracy for range queries by leveraging data structure (e.g., hierarchy, order, multi-dimensional spaces) while retaining the flexibility and unbiased estimation of traditional sampling. It achieves superior accuracy over structure-oblivious sampling and matches or exceeds tailored deterministic summaries like wavelets and q-digests, with minimal computational overhead and support for fast query processing.

ABSTRACT

In processing large quantities of data, a fundamental problem is to obtain a summary which supports approximate query answering. Random sampling yields flexible summaries which naturally support subset-sum queries with unbiased estimators and well-understood confidence bounds. Classic sample-based summaries, however, are designed for arbitrary subset queries and are oblivious to the structure in the set of keys. The particular structure, such as hierarchy, order, or product space (multi-dimensional), makes range queries much more relevant for most analysis of the data. Dedicated summarization algorithms for range-sum queries have also been extensively studied. They can outperform existing sampling schemes in terms of accuracy on range queries per summary size. Their accuracy, however, rapidly degrades when, as is often the case, the query spans multiple ranges. They are also less flexible - being targeted for range sum queries alone - and are often quite costly to build and use. In this paper we propose and evaluate variance optimal sampling schemes that are structure-aware. These summaries improve over the accuracy of existing structure-oblivious sampling schemes on range queries while retaining the benefits of sample-based summaries: flexible summaries, with high accuracy on both range queries and arbitrary subset queries.

Motivation & Objective

• To address the limitation of structure-oblivious sampling, which ignores inherent data structures like hierarchy, order, or multi-dimensional product spaces.
• To improve accuracy on range queries—common in real-world data analysis—without sacrificing the flexibility of sample-based summaries.
• To combine the benefits of random sampling (flexibility, unbiased estimators, exponential tail bounds) with the accuracy of deterministic summaries for range queries.
• To design a method that is both I/O efficient and scalable, requiring only two sequential passes over the data and independent memory usage from input size.

Proposed method

• Proposes variance-optimal sampling schemes that are explicitly designed to minimize estimation variance for range queries by incorporating structural knowledge of the data.
• Uses a two-pass algorithm: the first pass computes key statistics (e.g., weights, ranges), and the second pass selects samples based on structure-aware probabilities.
• Applies probability weighting proportional to the expected contribution of each key to range query variance, ensuring optimal variance reduction.
• Leverages hierarchical or multi-dimensional data structures (e.g., kd-trees, product spaces) to define range queries and guide sampling probabilities.
• Employs a modified Horvitz-Thompson estimator to ensure unbiased estimation with confidence bounds, even when sampling is structure-aware.
• Supports fast query answering by scanning the sample to compute intersections with query rectangles, enabling thousands of queries per second.

Experimental results

Research questions

• RQ1 Can sampling be made more accurate for range queries by incorporating structural knowledge of the data, without losing flexibility?
• RQ2 How does structure-aware sampling compare in accuracy and efficiency to both structure-oblivious sampling and deterministic summaries like q-digest and wavelets?
• RQ3 What is the trade-off between accuracy and computational cost when using structure-aware sampling versus wavelet-based or sketch-based summaries?
• RQ4 Can variance-optimality be preserved in a structure-aware context while maintaining the ability to handle arbitrary subset queries?

Key findings

• Structure-aware sampling achieves less than half the error of structure-oblivious sampling for sample sizes between 1% and 10% of the data, particularly on large-scale data with high-weight keys.
• For uniform area queries over 25 ranges, structure-aware sampling outperforms wavelets and q-digests, especially when the number of ranges increases.
• On the Tech Ticket dataset, wavelets required hours to build and query, while structure-aware sampling completed in seconds, highlighting the impracticality of wavelets for real-time analysis.
• Query processing with structure-aware samples is fast—thousands of rectangle queries per second—compared to wavelets, which take about a second per query due to decomposition into dyadic rectangles.
• The method maintains the same query time as structure-oblivious sampling, since both rely on intersection scans, but achieves significantly better accuracy.
• Building structure-aware samples requires only two sequential passes and independent memory usage, making it I/O efficient and scalable to large datasets.