star-dust-ctrl / deepsurv.pytorch

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},
}