LightGBM internals from scratch: XGBoost's boosting math, four tricks that make it fast — leaf-wise, histograms, GOSS, EFB
Article summary
Quick briefing — cleaned from the original RSS feed
XGBoost gave us a regularized second-order objective: a gradient gᵢ and Hessian hᵢ per sample, a closed-form leaf weight w* = −G/(H+λ) , and γ-gated gains. LightGBM keeps that exact math — every leaf is still −G/(H+λ) , every split scored by the same gain — and rebuilds the parts that make it fast on big data . I built a demo that computes all four accelerators for real, and once you see them the speedup stops being magic. Here they are. The shared core is unchanged Nothing about the statistics…
1Key Takeaways
- XGBoost gave us a regularized second-order objective: a gradient gᵢ and Hessian hᵢ per sample, a closed-form leaf weight w* = −G/(H+λ) , and γ-gated gains.
- LightGBM keeps that exact math — every leaf is still −G/(H+λ) , every split scored by the same gain — and rebuilds the parts that make it fast on big data .
- I built a demo that computes all four accelerators for real, and once you see them the speedup stops being magic.
- The shared core is unchanged Nothing about the statistics….
2AIWedia Score
8.3/10
High relevance — worth your attention today
Based on source trust, recency, category impact, and story depth.
3Why it matters
Coding AI shifts how fast software ships and how much human review each change needs. DEV — ML reports that xGBoost gave us a regularized second-order objective: a gradient gᵢ and Hessian hᵢ per sample, a closed-form leaf weight w* = −G/(H+λ) , and γ-gated gains.
Explore related
Browse toolsCoding AI news
Explore curated coding ai tools on AIWedia — compare, rank, and launch from our directory.
Full story on DEV — ML
Read full articleHeadlines aggregated via RSS for discovery on AIWedia. Original content © DEV — ML. We link to the source and do not republish full articles.