-
Squeezing more out of XGBoost: feature engineering, custom objectives, and constraints
Once learning rate, depth, regularization, subsampling, and early stopping are sensible, another search rarely creates a step-change. The next gains usually come from changing one of three things: