Skip to content

Dataset · stanford-star

Relational databases for RelBench v1 benchmark

stanford-star/relbench-v1

A collection of related parquet tables and task descriptions from Stanford for training neural networks on relational databases.

What's inside

The dataset provides a collection of relational databases and tasks from the first version of the RelBench benchmark by Stanford researchers (fewer than 1,000 top-level items). Each database includes parquet tables, schema links, manifest files with keys, timestamps for validation, and standard deviations for NMAE metric calculation.

What you can do with it

  • Training and comparing graph neural networks (GNNs) and tabular models on complex multi-table data
  • Testing prediction algorithms on relational databases with temporal splits
Size
up to 1K rows
Downloads this month
25.1K
Likes
6

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("stanford-star/relbench-v1")