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
- 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.
- Gradient boosting models:
- regression tree (GBRT)
- linear model/lasso
- Objectives to support tasks:
- regression
- classification
- OpenMP implementation
- More objective to support tasks:
- ranking
- matrix factorization
- structured prediction
- .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