Compact Multi-Class Boosted Trees

TL;DR

This paper introduces vector-valued trees and layer-by-layer boosting to create smaller, faster, and more interpretable multiclass gradient boosted models, especially suited for resource-constrained environments.

Contribution

The paper presents two novel extensions to gradient boosted decision trees: vector-valued trees for multiclass classification and layer-by-layer boosting for faster convergence and compact models.

Findings

Vector-valued trees reduce model size for multiclass problems.

Layer-by-layer boosting accelerates convergence and produces more compact ensembles.

Extensions improve performance on various multiclass datasets.

Abstract

Gradient boosted decision trees are a popular machine learning technique, in part because of their ability to give good accuracy with small models. We describe two extensions to the standard tree boosting algorithm designed to increase this advantage. The first improvement extends the boosting formalism from scalar-valued trees to vector-valued trees. This allows individual trees to be used as multiclass classifiers, rather than requiring one tree per class, and drastically reduces the model size required for multiclass problems. We also show that some other popular vector-valued gradient boosted trees modifications fit into this formulation and can be easily obtained in our implementation. The second extension, layer-by-layer boosting, takes smaller steps in function space, which is empirically shown to lead to a faster convergence and to a more compact ensemble. We have added both improvements to the open-source TensorFlow Boosted trees (TFBT) package, and we demonstrate their efficacy on a variety of multiclass datasets. We expect these extensions will be of particular interest to boosted tree applications that require small models, such as embedded devices, applications requiring fast inference, or applications desiring more interpretable models.