license: mit
task_categories:
- graph-ml
RandSATBench: Benchmarks for Constraint Satisfaction Problems
This repository contains the benchmarks presented in the paper Benchmarking Graph Neural Networks in Solving Hard Constraint Satisfaction Problems.
RandSATBench provides a mixed set of easy, hard, and unsatisfiable instances of constraint satisfaction problems, specifically q-coloring and Boolean satisfiability (K-SAT) problems. The goal is to provide a challenging setting to compare the performance of deep learning methods (particularly Graph Neural Networks) against classical exact and heuristic solvers.
- Paper: https://huggingface.co/papers/2602.18419
- Repository: https://github.com/ArtLabBocconi/RandCSPBench
Datasets
The benchmark includes datasets for Boolean satisfiability (3-SAT and 4-SAT) and coloring problems (3-coloring and 5-coloring). The instances vary in size ($N=16, 32, 64, 256$) and connectivity.
| Dataset | # Train Instances | # Test Instances |
|---|---|---|
| 3-SAT | 168,000 | 42,000 |
| 4-SAT | 84,000 | 21,000 |
| 3-col | 60,000 | 20,000 |
| 5-col | 60,000 | 20,000 |
Usage
K-SAT
The random instances of 3-SAT or 5-SAT problems in CNF format can be downloaded using the following:
wget https://huggingface.co/datasets/CarloLucibello/kSAT-Benchmarks/resolve/main/kSAT.zip
unzip kSAT.zip
Ground truth solutions obtained by running the CaDiCal solver are available in the corresponding train_labels.csv files.
q-coloring
To generate the random graphs for the 3-coloring and 5-coloring benchmarks, use the scripts provided in the GitHub repository:
cd datasets
python gen_graphs_coloring.py
Citation
If you use this benchmark in your research, please cite:
@article{lucibello2026benchmarking,
title={Benchmarking Graph Neural Networks in Solving Hard Constraint Satisfaction Problems},
author={Lucibello, Carlo and others},
journal={arXiv preprint arXiv:2602.18419},
year={2026}
}