Skip to content

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

steb all_to_all_pair_classification rrivera1849/LUAR-MUD --dataset corpus-of-diverse-styles -e 5

Datasets

List available datasets:

steb all_to_all_pair_classification --list-datasets