xgboost: eXtreme Gradient Boosting

A General purpose gradient boosting (tree) library.

Authors:

• Tianqi Chen, project creater
• Kailong Chen, contributes regression module

Turorial and Documentation: https://github.com/tqchen/xgboost/wiki

Features

• Sparse feature format:
  • Sparse feature format allows easy handling of missing values, and improve computation efficiency.
• Push the limit on single machine:
  • Efficient implementation that optimizes memory and computation.
• Layout of gradient boosting algorithm to support generic tasks, see project wiki.

Supported key components

• Gradient boosting models:
  • regression tree (GBRT)
  • linear model/lasso
• Objectives to support tasks:
  • regression
  • classification
• OpenMP implementation

Planned components

• More objective to support tasks:
  • ranking
  • matrix factorization
  • structured prediction

File extension convention

• .h are interface, utils and data structures, with detailed comment;
• .cpp are implementations that will be compiled, with less comment;
• .hpp are implementations that will be included by .cpp, with less comment