Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Benchmarking RL-Enhanced Spatial Indices Against Traditional, Advanced, and Learned Counterparts

Guanli Liu, Renata Borovica-Gajic, Hai Lan, Zhifeng Bao

TL;DR

This paper presents the first modular benchmark framework for RL-enhanced spatial indices (RLESIs), designed to compare RLESIs with traditional, advanced, and learned spatial indices across diverse datasets and workloads. The framework separates index training (ITM) from index building (IBM) and uses grid-search parameter tuning to optimize performance under build-time constraints. Through evaluations on six datasets and twelve indices, the study finds that while RLESIs can reduce query latency with tuning, they generally underperform learned spatial indices and advanced baselines, and they incur substantial training and build-time costs. The work identifies training cost, parameter sensitivity, and generalization gaps as key barriers, and suggests improvements such as cost-based reward functions and incremental learning to make RLESIs more practical in real systems.

Abstract

Reinforcement learning has recently been used to enhance index structures, giving rise to reinforcement learning-enhanced spatial indices (RLESIs) that aim to improve query efficiency during index construction. However, their practical benefits remain unclear due to the lack of unified implementations and comprehensive evaluations, especially in disk-based settings. We present the first modular and extensible benchmark for RLESIs. Built on top of an existing spatial index library, our framework decouples index training from building, supports parameter tuning, and enables consistent comparison with traditional, advanced, and learned spatial indices. We evaluate 12 representative spatial indices across six datasets and diverse workloads, including point, range, kNN, spatial join, and mixed read/write queries. Using latency, I/O, and index statistics as metrics, we find that while RLESIs can reduce query latency with tuning, they consistently underperform learned spatial indices and advanced variants in both query efficiency and index build cost. These findings highlight that although RLESIs offer promising architectural compatibility, their high tuning costs and limited generalization hinder practical adoption.

Benchmarking RL-Enhanced Spatial Indices Against Traditional, Advanced, and Learned Counterparts

TL;DR

This paper presents the first modular benchmark framework for RL-enhanced spatial indices (RLESIs), designed to compare RLESIs with traditional, advanced, and learned spatial indices across diverse datasets and workloads. The framework separates index training (ITM) from index building (IBM) and uses grid-search parameter tuning to optimize performance under build-time constraints. Through evaluations on six datasets and twelve indices, the study finds that while RLESIs can reduce query latency with tuning, they generally underperform learned spatial indices and advanced baselines, and they incur substantial training and build-time costs. The work identifies training cost, parameter sensitivity, and generalization gaps as key barriers, and suggests improvements such as cost-based reward functions and incremental learning to make RLESIs more practical in real systems.

Abstract

Reinforcement learning has recently been used to enhance index structures, giving rise to reinforcement learning-enhanced spatial indices (RLESIs) that aim to improve query efficiency during index construction. However, their practical benefits remain unclear due to the lack of unified implementations and comprehensive evaluations, especially in disk-based settings. We present the first modular and extensible benchmark for RLESIs. Built on top of an existing spatial index library, our framework decouples index training from building, supports parameter tuning, and enables consistent comparison with traditional, advanced, and learned spatial indices. We evaluate 12 representative spatial indices across six datasets and diverse workloads, including point, range, kNN, spatial join, and mixed read/write queries. Using latency, I/O, and index statistics as metrics, we find that while RLESIs can reduce query latency with tuning, they consistently underperform learned spatial indices and advanced variants in both query efficiency and index build cost. These findings highlight that although RLESIs offer promising architectural compatibility, their high tuning costs and limited generalization hinder practical adoption.

Paper Structure

This paper contains 45 sections, 4 equations, 24 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Spatial Queries and Indices Revisited
  3. Spatial Queries
  4. Classification of Spatial Indices
  5. Data Partitioning-based (DP-based)
  6. Space Partitioning-based (SP-based)
  7. Mapping-based (MP-based)
  8. Benchmarking Framework
  9. Implementation of Parameter Tuning
  10. Implementation of Index Building
  11. Index Nodes
  12. Insertion Strategy
  13. Selection of Baselines
  14. Experimental Setup
  15. Datasets
  16. ...and 30 more sections

Figures (24)

  • Figure 1: An overview of the benchmarking framework.
  • Figure 2: An overview of RLR-tree build.
  • Figure 3: An overview of Qd-tree build.
  • Figure 4: An example of a piecewise SFC (BM-tree).
  • Figure 5: Framework for benchmarking spatial indices: learning, deployment, and build in libspatialindex.
  • ...and 19 more figures

Theorems & Definitions (4)

  • Definition 1: Point Query (PQ)
  • Definition 2: Range Query (RQ)
  • Definition 3: $k$ Nearest-Neighbor Query ($k$NNQ)
  • Definition 4: Spatial Join (SJ)