All-to-All Pair Classification¶
Evaluates whether embeddings can distinguish same-class vs. different-class text groups using cosine similarity across all pairs.
Metrics¶
- EER (Equal Error Rate): lower is better
- AUC: higher is better
- AUC@FPR: AUC at false positive rate thresholds 0.01, 0.05, 0.10, 0.20, 0.30, 0.50
CLI¶
Datasets¶
List available datasets: