Boosted Trees

In this chapter, we describe boosted trees: the TensorFlow (TF) approach to gradient boosting. It is a class of ML algorithms that produce a prediction model in the form of an ensemble of weak prediction models, typically decision trees. The model is constructed in a stage-wise fashion and generalized by utilizing an arbitrary (differentiable) loss function. Gradient boosted trees are an extremely popular class of algorithms, as they can be parallelized (at the tree construction stage), can natively handle missing values and outliers, and require minimal data preprocessing.