TF Boosted Trees: A scalable TensorFlow based framework for gradient boosting

摘要

TF Boosted Trees (TFBT) is a new open-sourced frame-work for the distributedtraining of gradient boosted trees. It is based on TensorFlow, and itsdistinguishing features include a novel architecture, automatic lossdifferentiation, layer-by-layer boosting that results in smaller ensembles andfaster prediction, principled multi-class handling, and a number ofregularization techniques to prevent overfitting.