DeepSurv.pytorch

This repository is an unofficial pytorch implementation of DeepSurv: Personalized Treatment Recommender System Using A Cox Proportional Hazards Deep Neural Network. We reimplement the experiments in the paper, which is followed by Github, and the detailed understanding is available on my Blog.

Requirements

  • Pytorch>=0.4.0
  • CPU or GPU
  • Other packages can be installed with the following instruction:
pip install requirements.txt

Quick start

Running the code with the following command.

python main.py

Note: You can modify some parameters in configs/*.ini to get your own specific models.

Results

Simulated Linear Simulated Nonlinear WHAS SUPPORT METABRIC Simulated Treatment Rotterdam & GBSG
Paper 0.774019 0.648902 0.862620 0.618308 0.643374 0.582774 0.668402
Our implements 0.778607 0.652048 0.841484 0.618107 0.643453 0.552648 0.673290

Citation

@article{Katzman2016DeepSurv,
  title={DeepSurv: Personalized Treatment Recommender System Using A Cox Proportional Hazards Deep Neural Network},
  author={Katzman, Jared and Shaham, Uri and Bates, Jonathan and Cloninger, Alexander and Jiang, Tingting and Kluger, Yuval},
  journal={Bmc Medical Research Methodology},
  volume={18},
  number={1},
  pages={24},
  year={2016},
}