Concepts¶
Understanding the algorithms that power Compere.
Overview¶
Compere uses two algorithms to create an efficient ranking system:
- Elo Rating System - Updates entity ratings after each comparison
- Multi-Armed Bandit (UCB) - Selects optimal pairs for comparison
Contents¶
- Elo Rating System - How ratings are calculated and updated
- MAB Algorithm - Upper Confidence Bound for pair selection
- Entity Pairing - How Compere selects comparison pairs
Algorithm Selection Guide¶
| Scenario | Recommended Approach |
|---|---|
| Quick ranking with few comparisons | MAB/UCB (exploration-focused) |
| Stable, established rankings | Lower K-factor (16-24) |
| New entities added frequently | Default settings (K=32, UCB with cold-start) |
| Diverse comparison coverage | Similarity matching or increase random factor |
| Statistical validity required | Random pairing (no UCB) |