This repository allows to reproduce the results in Neural Survival Clustering: Non parametric mixture of neural networks for survival clustering This paper approaches survival analysis as a mixture of neural baselines whereby different baseline cumulative hazard functions are modelled using positive and monotone neural networks.
The model consists in a mixture of K neural networks that model different population-level survival distributions, and an assignment neural network that weights these different mixtures given the input data.
To use the model, one needs to execute:
from nsc import NeuralSurvivalClustering
model = NeuralSurvivalClustering()
model.fit(x, t, e) # Fit the model on covariates x, with event time t, and event indicator e (e = 1 means event)
model.predict_survival(x, horizon) # Predict the survival for covariates x at a given horizon
To then explore the clustering structre:
model.predict_alphas(x) # Obtain the soft assignemnt to the different cluster
model.survival_cluster(horizons) # Compute the survival of the different clusters at different horizons
A full example with analysis is provided in examples/Neural Survival Clustering on SUPPORT Dataset.ipynb
.
To reproduce the paper's results:
- Clone the repository with dependencies:
git clone [email protected]:Jeanselme/NeuralSurvivalClustering.git --recursive
- Create a conda environment with all necessary libraries
pycox
,lifelines
,pysurvival
. - Add path
export PYTHONPATH="$PWD:$PWD/DeepSurvivalMachines:$PWD/SuMo-net:$PYTHONPATH"
- Run
examples/experiments_paper.py METABRIC
- Repeat with
SUPPORT
andSYNTHETIC
- Analysis using
examples/Analysis.ipynb
Adding a new method consists in adding a child to Experiment
in experiment.py
with functions to compute the nll and fit the model.
Then, add the method in examples/experiments_paper.py
and follow the previous point.
TODOs
have been added to make the addition of a new method easier.
- Competing risks.
- Post analysis: Replacing assignment neural network by more interpretable approach.
We followed the same structure than the DeepSurvivalMachines repository with the model in nsc/
- only the api should be used to test the model. Examples are provided in examples/
.
git clone [email protected]:Jeanselme/NeuralSurvivalClustering.git --recursive
The model relies on DeepSurvivalMachines
, SuMo-net
, pytorch
, numpy
and tqdm
.
To run the set of experiments pycox
, lifelines
, pysurvival
are necessary.