This repository contains the code accompanying the paper "Uncertainty-Aware Benefit Estimation: A Reliable Framework for Index Tuning." which is currently under submission to VLDB 2025.
The folder index_selection_evaluation/ML/AImeetsAI includes various uncertainty-aware models based on the paper "AI Meets AI: Leveraging Query Executions to Improve Index Recommendations." These models include:
- AMA-BNN
- AMA-MCD-all
- AMA-MCD-last
- AMA-MCD-dpp
- AMA-Ensemble
- AMA-AE
- AMA-AE-MCD
The script index_selection_evaluation/selection/ProcessedWorkloadBeauty.py demonstrates how to integrate the BE-UQ models into the index tuning process.
- Uncertainty-Aware Benefit Estimation: A Reliable Model for Index Tuning
- Refactoring Index Tuning Process with Benefit Estimation
- AI Meets AI: Leveraging Query Executions to Improve Index Recommendations
- Magic mirror in my hand, which is the best in the land?: an experimental evaluation of index selection algorithms
For detailed usage and implementation, please refer to the individual scripts and the accompanying documentation within the repository.