ILP Datasets: 55 Relational Learning and Inductive Logic Programming Collections

A collection of 55 relational learning and inductive logic programming datasets curated by andrewcropper. The datasets span multiple domains, including visual reasoning and molecular biology, with specific tasks like 1D-ARC grid transformations and antiviral activity prediction. The dataset page was last updated on 2026-06-17.

Use Cases

• Training models for visual reasoning tasks based on the 1D-ARC grid transformation tasks.
• Developing predictive models for drug design based on molecular data for Alzheimer's disease.
• Benchmarking relational learning algorithms on antiviral activity prediction tasks against HIV.
• Evaluating inductive logic programming systems across a diverse set of 55 logical and relational problems.

Strengths

• Contains 55 distinct datasets, providing a broad testbed for relational learning.
• Includes domain-specific datasets like 'alzheimer' for drug design, with source attribution to King et al. 1995.
• Features structured tasks, such as '1d' with 18 types × 50 instances of visual reasoning problems.

Limitations

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

Provenance

Source

Hugging Face user andrewcropper.

Collection Method

Curated collection from various academic sources.

Freshness

Last updated 2026-06-17 13:26:12; freshness should be verified.