🐹

【ML】GBDT Libraries Summay

2024/08/13に公開

0. Summary of GBDT Libraries

There are three popular libraries in GBDT, LGBM, CatBoost, XGBoost.
This time, I summarize the features of each libraries briefly, the detail explanation is written in another article, so please check it when you need.

1. LGBM

Features
・Fast
・Accurate
・Memory Efficiency

Article
Detail Explanation with code.
https://zenn.dev/yuto_mo/articles/b0843b74733688

2. CatBoost

Features
・Can handle categorical features as is.
・Less overfitting
・Robust to hyperparameters

Article
Detail Explanation with code.
https://zenn.dev/yuto_mo/articles/d04b24fd87d272

3. XGBoost

Features
・Great scalability and flexibility.

Article
Detail Explanation with code.
https://zenn.dev/yuto_mo/articles/a7dd1aec3356f3

4. In the end

The ensemble of models generally performs a greater score than one model. So, each library has each feature, please consider using all as models that are to be ensembled.

・Ensemble method
https://zenn.dev/yuto_mo/articles/3a7a8237c700e0

In my experience, the blending works in following order;
①good and highly different models > ②not good but highly different models ≒ ③good and slightly different models >> ④not good and not different models.
quote: [3]

Reference

[1] https://qiita.com/Ak_ki/items/70572ae1c1728335219e
[2] https://qiita.com/Ak_ki/items/38b85992547454005533
[3] https://www.kaggle.com/competitions/open-problems-multimodal/discussion/363230

Discussion