Ensemble Blend
Combines multiple submission CSVs into one blended submission. Rank averaging is robust to scale differences; weighted averaging preserves magnitude.
When to use
- You have 2+ models with similar CV scores but different architectures (diversity → blend gain).
- CV/LB correlation is strong enough that you trust individual submissions.
- You want to reduce variance before a final submission.
Rule of thumb
| Situation | Method |
|---|---|
| Regression, same scale | Weigh |
[Description truncada. Veja o README completo no GitHub.]