Purpose
Classify scientific objects and detect patterns using established taxonomies and classification schemes.
Key Datasets
- SDSS Stellar Classification (Allanatrix/Astro): 100K objects from SDSS DR17 — Stars, Galaxies, Quasars with photometric features (u, g, r, i, z magnitudes, redshift)
- Social Bias Frames (allenai/social_bias_frames): Allen AI SBIC corpus for detecting implicit social biases in text
Protocol
- Feature extraction — Identify relevant features for classification task
- Taxonomy mapping — Map to standard classification scheme
- Classification — Apply appropriate classifier with confidence scores
- Validation — Cross-validate against known labeled examples
- Edge case analysis — Flag ambiguous or borderline cases
Classification Domains
- Astronomical objects: Stellar spectral types (OBAFGKM), galaxy morphology (Hubble), AGN types
- Biological taxonomy: Species classification, protein families, cell types
- Chemical compounds: Functional groups, drug classes, toxicity levels
- Text classification: Sentiment, bias detection, topic classification
- Image classification: Histopathology, satellite imagery, microscopy
Rules
- Report classification confidence and alternative labels
- Use domain-standard taxonomies (not ad-hoc categories)
- Handle multi-label and hierarchical classification
- Document decision boundaries and feature importance