Home

Introduction

XGBoost (short for eXtreme Gradient Boosting) is an efficient general purpose gradient boosting library. Via easy configuration, we can use different boosting models and objective functions to fit the real world data. To get started, read Binary Classification example.

Examples

• Binary Classification (read this for quick start)
• Regression
• Ranking
• See more examples in the demo folder

Informations

• Input Data Format
• Python Interface

Questions about the toolkit

If you have questions, suggestions, fire an issue

Boosters

• Tree Booster
• Linear Booster

Parameter Setting

• Detailed parameter settings are provided in Parameters

Compile XGBoost

• To compile, simply type make
• In clang(common in Mac machines), you may get an error called -lgomp is not found. This is because your compiler does not have openmp support(while linker seems to disagree..)
  • You can simply remove -fopenmp flag in Makefile, this will give you single thread version of xgboost
  • You can also try to update your compiler to get OpenMP support

Project Layout

The project have three layers, the listed files are commented headers that are useful to use code these layers.

• Booster core interface: booster/xgboost.h, booster/xgboost_data.h.
  • Provides a interface to a single gradient boosters, all implements hides behind this interface
  • Use this interface to add new implementation boosters, or use create booster to do specific tasks
• Booster ensemble base class: booster/xgboost_gbmbase.h
  • Provides a base class that provides useful code for booster ensemble, provides the buffering scheme.
  • Use this class to create customized learner with self-defined loss function. Take class GBMBase, use Predict to get predictions and calculate gradient and second order gradient, put the statistics back to DoBoost to update the model.
• Booster task wrappers: regression, rank(beta)
  • Provides direct wrapper to do specific learning tasks