Bridging Machine Learning and Query Optimization: Feedback-Driven Selectivity Estimation for Spatial Filters

This repository contains code and resources related to our research on feedback-driven spatial selectivity estimation. The project focuses on leveraging optimizer feedback to improve the estimation of selectivity for multi-dimensional spatial predicates. Various Machine Learning models, including neural networks, tree-based models, and instance-based models, are explored to address this challenging task efficiently across different spatial filter types.

Code Structure

The repository is organized as follows:

• analyse_results: Contains all code for generating figures, plots, and conducting statistical tests presented in our study
• intersect_filter: Implementation of our ML approach for intersect-type spatial selectivity estimation
• contain_filter: Implementation of our ML approach for containment-type spatial selectivity estimation
• distance_filter: Implementation of our ML approach for distance-based spatial selectivity estimation
• traditional_methods: Implementation of baseline approaches (RTree and Histogram-based estimation) used for comparison

Additional Resources

To facilitate reproduction of our results without requiring lengthy retraining of models, we provide the following direct downloads:

• datasets_intersects.zip (2.47 GB): contains feedback on the 14 datasets for the INTERSECTS spatial filter.
• datasets_contains.zip (2.47 GB): contains feedback on the 14 datasets for the CONTAINS spatial filter.
• datasets_distance.zip (1.28 GB): contains feedback on the 14 datasets for the DISTANCE spatial filter.
• spatial_dump_file.dump (6.62 GB): PostgreSQL dump of all 14 spatial databases used in our experiments.
• traditional_methods.zip (3.81 GB): build artifacts, data, and configurations for the histogram- and RTree-based baseline estimators.
• learned_models.zip (45.46 GB): saved learned models to shorten setup time and ensure reproducibility of the reported results.

License

The work in this repository is licensed under the MIT License. Please refer to the LICENSE file for more details.

Contributors

1. Nadir GUERMOUDI (LIAS/ISAE-ENSMA & LRIT/University of Tlemcen)
2. Houcine MATALLAH (LRIT/University of Tlemcen)
3. Amin MESMOUDI (LIAS/University of Poitiers)
4. Seif-Eddine BENKABOU (LIAS/University of Poitiers)
5. Allel HADJALI (LIAS/ISAE-ENSMA)
6. Ahmed-Youcef BENHALIMA (LIAS/ISAE-ENSMA)