STRABLE Benchmark: Tabular Machine Learning with Strings

The STRABLE benchmark is a suite for evaluating machine learning models on tabular data containing strings, addressing a previously understudied setting. It was created by inria-soda and is hosted on Hugging Face. The dataset was last updated on June 11, 2026.

Use Cases

• Benchmarking model performance on tabular data containing strings.
• Evaluating dedicated architectures for heterogeneous tabular data.
• Comparing machine learning methods on real-world tables with non-numeric entries.

Strengths

• Designed specifically for benchmarking tabular ML with strings, a noted gap.
• Created by a recognized research organization (inria-soda).

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

inria-soda

Freshness

Last updated 2026-06-11 14:03:49.