A dataset created by Kairong-Han and last updated on November 2, 2025, designed to evaluate machine learning models under spurious correlations. It contains two subtasks, STG_E and STG_H, each with training and test splits including in-distribution (IID) and out-of-distribution (OOD) data. The STG_E subtask includes three training splits of varying sizes or difficulty levels.
Use Cases
- Benchmarking model robustness based on the described IID and OOD test splits
- Evaluating model performance on spurious correlations based on the dataset's stated purpose
- Comparing model scaling effects based on the STG_E subtask's multiple training splits of different sizes
- Studying out-of-distribution generalization based on the dedicated OOD test splits
Strengths
- Dataset is structured with two distinct subtasks (STG_E and STG_H) for focused evaluation
- STG_E subtask provides three training splits (STG_S, STG_M, STG_L) representing different data sizes or difficulty levels
- Each subtask includes dedicated in-distribution (IID) and out-of-distribution (OOD) test splits
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
- huggingface
- Freshness
- Last updated 2025-11-02 07:41:47; freshness should be verified