GBST

GBST is an optimized distributed gradient boosting survival trees library that is implemented based on the XGBoost(https://github.com/dmlc/xgboost). It is designed to be highly efficient, flexible and portable. It implements Gradient Boosting Survival Trees algorithms from the paper Gradient Boosting Survival Tree with Applications in Credit Scoring. With XGBoost as its core, GBST inherits the parallel tree boosting (also known as GBDT/GBM) to solve the problem of survival analysis efficiently and accurately .

License

© Contributors, 2019. Licensed under an Apache-2 license.

Installation

• Requirements

  • (for python package) python3(>=3.5.0), numpy, scipy, scikit-learn, (optional) pandas, (optional) matplotlib.
  • (for compiling from source) All the above, plus CMake(>=3.12), and a C++ compiler supporting C++11(e.g. g++-5.0 or higher)

• Install the python package only We provide the compiled library file (for Win64 and Linux) within the python package. To install, simply execute

  python setup.py install
  

  in the repository "./gbst_package".

• Compile from source The folder gbst_src contains the source codes to compile the library file. To build it from source, we start from making a working directory:

  cd gbst_src
  mkdir build
  

  Then, for Windows:

  cmake .. -G "Visual Studio 14 2015 Win64"
  # for VS15: cmake .. -G"Visual Studio 15 2017" -A x64
  # for VS16: cmake .. -G"Visual Studio 16 2019" -A x64
  cmake --build . --config Release
  

  If the build process successfully ends, you should find the library file gbst.dll in the ./gbst_src/lib folder.

  For linux:

  cmake .. 
  make -j$(nproc)
  

  If the build process successfully ends, you should find the library file libgbst.so in the ./gbst_src/lib folder.

  After that, please copy the compiled library into gbst_package/GBST/ , in place of the pre-compiled library file(s), then install the python package by previous instructions.

Usage

As in XGBoost, GBST provides interfaces to scikit-learn. One can easily define and tune a Survival Tree model, save/load it via local storage devices, and infer the hazard functions by the model.

• Interfaces to scikit-learn, inherit from sklearn::BaseEstimator
• Hyper parameter searching/cross validation, as is in sklearn
• Plotting the tree's structure in Graphviz
• Plotting the importance of each feature

Reference

• Miaojun Bai, Yan Zheng, and Yun Shen. Gradient Boosting Survival Tree with Applications in Credit Scoring. arXiv preprint arXiv:1908.03385 (2019).
• Tianqi Chen and Carlos Guestrin. XGBoost: A Scalable Tree Boosting System. In 22nd SIGKDD Conference on Knowledge Discovery and Data Mining, 2016