-
Debugging a boosted model: learning curves, tree dumps, and plots that explain a prediction
A boosted-tree model can post an attractive validation score while still being wrong for the reasons that matter: leakage, an unrepresentative split, a brittle interaction, or one small segment carryi