Voting, Stacking, and Blending Ensembles
The premise is that models failing differently cover for each other. It is often false.
-
-
XGBoost, LightGBM, and CatBoost in Practice
The same algorithm, three engineering positions. On numeric data you cannot tell them apart.
-
Gradient Boosting from First Principles
Boosting fits successive trees to the current ensemble’s loss derivatives. Its behavior depends on the loss, step size, and fitting choices.
-
Bagging, Random Forests, and Extra Trees
Three answers to one question: how to make the members disagree without making them worse.
-
Decision Tree Splits, Pruning, and Instability
Readable, scale-free, and unstable enough that one tree is rarely the answer.