LightGBM
Stub. Created as a placeholder — expand with vault-sourced content.
What it is
Microsoft’s gradient-boosted decision tree library. Its lambdarank objective implements LambdaMART, making it a common choice for feature-based LTR rerankers. Exposes ndcg_eval_at and uses the Burges (exponential-gain) NDCG variant by default.
Vault references (existing coverage)
- Learning to Rank —
LGBMRanker(objective="lambdarank", ...)code snippet - Building a Better Search Engine for Semantic Scholar — LightGBM + LambdaRank reranker, 22 engineered features
- NDCG / Flavors of NDCG — LightGBM uses the Burges NDCG variant
- Hybrid Search and Learning-to-Rank with Metarank — LightGBM as a LambdaMART implementation; missing-value handling
TODO
- Document ranking objectives and how they map to Ranking Objectives.