Leaf Tax

XGBoost's objective function adds a regularization term that penalizes both the number of leaves (weighted by gamma) and the size of leaf weights (an L2 penalty weighted by lambda), and it speeds up training with histogram-based split finding that buckets continuous feature values into discrete bins before scanning for the best split. Which statement correctly describes gamma's role in this objective?

Gamma is XGBoost's learning rate applied to each boosting round's tree output

Gamma is the minimum loss reduction required to justify a further split, so raising it prunes more leaves and produces smaller trees

Gamma controls the number of histogram bins used when searching for candidate split points

Gamma shrinks the leaf weights themselves via an L2 penalty, independent of tree structure