DBinfer: Relational Datasets for Machine Learning Benchmarking

RelBench hosts the dbinfer family of relational datasets in its 3.0 manifest format. The datasets originate from the 4DBInfer benchmark and are exposed via the dbinfer-relbench-adapter package. Labels for tasks are built externally and served as-is.

Use Cases

• Benchmarking relational machine learning models based on the 4DBInfer benchmark structure.
• Evaluating graph neural network performance on tasks with externally provided labels.
• Testing data processing pipelines using the RelBench 3.0 manifest and Parquet file format.

Strengths

• Datasets follow a standardized RelBench 3.0 manifest format for consistency.
• Data originates from the established 4DBInfer benchmark, suggesting a research foundation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

4DBInfer benchmark (data version 20240304)

Collection Method

Exposed to RelBench via the dbinfer-relbench-adapter package.

Freshness

Last updated 2026-06-12 07:58:38; freshness should be verified.