Competitive Analysis of the Top Gradient...
summary
Comparison measures four GBDT frameworks (XGBoost, CatBoost, LightGBM, SnapBoost) on Pokemon numeric and categorical, CIFAR‑10 image, and Melbourne daily temps using HyperOpt tuning on an IBM PowerAI AC922 CPU; results show SnapBoost hits 99% on the numeric set and is the fastest on AC922 after HPO, CatBoost scores 61% on CIFAR‑10 but is often slow, LightGBM is the quickest for most datasets except images, and XGBoost gains ~13% from HPO while temporal accuracy improves up to 27% for SnapBoost.
AI-written summary · about
chapters (20)
- 0:00Introduction and acknowledgments
- 1:02Why gradient boosting matters
- 2:25Gradient boosting explained
- 3:47Research methodology overview
- 4:50Runtime environment details
- 5:10Datasets used in analysis
- 6:36Algorithm competitors overview
- 6:57XGBoost features explained
- 7:59LightGBM features explained
- 8:40CatBoost features explained
- 9:22SnapBoost features explained
- 10:02Hyperparameter optimization
- 11:05Experimental results overview
- 12:08Numeric dataset performance
- 13:11Categorical dataset performance
- 14:13Temporal dataset performance
- 15:18Image dataset performance
- 16:01Key findings and inferences
- 17:24Impact of hyperparameter tuning
- 17:45Conclusion and future work
related talks
38mCatBoost - the new generation of gradient boosting
01h 27mCatBoost - The New Generation of Gradient Boosting
01h 33mCrunching Your Data with CatBoost - the New Gradient Boosting Library
01h 28mMastering gradient boosting with CatBoost
41mBetter Than Deep Learning: Gradient Boosting Machines (GBM)
37mUnlocking the Power of Gradient-Boosted Trees (using LightGBM)
32mDeep Dive into scikit-learn's HistGradientBoosting Classifier..
40mCatBoost: Fast Open-Source Gradient Boosting Library For GPU