This project is targeted for survival prediction on cardivascular data.
To run the codes, please put "LRPP_updated.csv" in data folder. Run data.py and data.pkl will be generated.
This folder treats a survival problem as different classification problems. Unclear labels, i.e., censoring before targeting time is deleted. The purpose is to compare the performance difference of neural network and logistic regression on the dataset.
classification/logistic.py : logistic regression classification/nn.py: neural network to perform classification
This folder is implementation of https://peerj.com/articles/6257/ In this framework, survival times are discretized into several buckets. And loglikelihood is used as loss function.
dsn/dsn2.py: discretize time into two buckets. This is same as classifcation problem but take censoring into account dsn/dsn5.py: discretize time into five buckets. This aligns with original implementation in the paper dsn/dsnfull.py: discretize time by year. This is the most useful survival model since it predicts survival probability every year.
common.py: all definition of feature lists (MARKERS, BASE_COVS etc) data.py: generation of data from original lrpp_updated.csv loss.py: all loss functions and evaluation metric as c-index and aucJM models.py: all definition of neural networks preprocess.py: summarize features of original dataset r_connection.py: compare aucJM and c-index function from exported R results utils.py: some utility functions
This is a depreciated folder which is an (unsuccessful) implementation of https://arxiv.org/pdf/1705.10245.pdf