RelBench

Website | Vision Paper | Mailing List |

Relational Deep Learning is a new approach for end-to-end representation learning on data spread across multiple tables, such as in a relational database (see our vision paper). RelBench is the accompanying benchmark which seeks to facilitate efficient, robust and reproducible research in this direction. It comprises of a collection of realistic, large-scale, and diverse datasets structured as relational tables, along with machine learning tasks defined on them. It provides full support for data downloading, task specification and standardized evaluation in an ML-framework-agnostic manner. Additionally, there is seamless integration with PyTorch Geometric to load the data as a graph and train GNN models, and with PyTorch Frame to encode the various types of table columns. Finally, there is a leaderboard for tracking progress.