BEACON

Benchmark for Efficient and Accurate Counting of Subgraphs

BEACON is the first comprehensive benchmark for subgraph counting, enabling systematic comparisons of algorithmic (AL) and machine learning (ML) methods. It provides standardized datasets, reproducible evaluation protocols, and a public leaderboard to accelerate research in graph analytics.

26,435 Graphs
Subgraphs up to 5 nodes
10+ Evaluated Methods
Subgraph counting example visualization

Key Features

BEACON introduces several innovations to the field of subgraph counting

Standardized Datasets
Curated Oracle Dataset with verified ground truths

Our Oracle Dataset contains 26,435 graphs from TUDataset and OGB, covering bioinformatics, social networks, and more.

Unified Evaluation
Compare AL and ML methods under identical conditions

BEACON enables direct comparison between algorithmic methods (e.g., ESCAPE) and machine learning approaches (e.g., PPGN).

BEACON-Sampler
Generate custom datasets by node count, density, and domain

Our PyPI tool allows researchers to create custom graph datasets with specific properties for targeted evaluation.

Public Leaderboard
Track state-of-the-art performance across metrics

Our leaderboard tracks performance across accuracy (Q-error, MAE), scalability, and robustness metrics.

Latest Updates

Stay informed about the latest developments in our benchmark

Version History

Jun 2024

BEACON v1.2 Release

Added new network graphs

Feb 2024

BEACON v1.1 Release

Improved evaluation server and documentation

Top Submissions

PPGN (ML)

Q-error: 1.02

Dataset: Set_1

Type: Machine Learning

DeSCo (ML)

Q-error: 1.07

Dataset: Set_3

Type: Machine Learning

ESCAPE (AL)

Q-error: 1.23

Dataset: Set_1

Type: Algorithmic