Treelite: Toolbox for Decision Tree Deployment

This paper introduces a brand new tree library treelite1. The library is a toolbox to facilitate easy deployment of models and accelerate prediction performance. It has a Python wrapper that allows users to integrate it as part of their workflow. Treelite is able to read tree ensemble models that are trained by any tree libraries, including XGBoost [1], LightGBM [2], and scikit-learn [3]. Treelite is also designed to minimize dependencies at the time of deployment. It used to be the case that one had to ship his tree model with the original tree library that trained it; with treelite, it is no longer. Finally, treelite allows for optimizations that improve prediction performance without changing any detail of the model.