Package dsm

Python package dsm provides an API to train the Deep Survival Machines and associated models for problems in survival analysis. The underlying model is implemented in pytorch.

What is Survival Analysis?

Survival Analysis involves estimating when an event of interest, $T$ would take places given some features or covariates $X$. In statistics and ML these scenarious are modelled as regression to estimate the conditional survival distribution, $\mathbb{P}(T>t|X)$. As compared to typical regression problems, Survival Analysis differs in two major ways:

• The Event distribution, $T$ has positive support ie. $T \in [0, \infty)$.
• There is presence of censoring ie. a large number of instances of data are lost to follow up.

Deep Survival Machines

Deep Survival Machines (DSM) is a fully parametric approach to model Time-to-Event outcomes in the presence of Censoring first introduced in [1]. In the context of Healthcare ML and Biostatistics, this is known as 'Survival Analysis'. The key idea behind Deep Survival Machines is to model the underlying event outcome distribution as a mixure of some fixed $k$ parametric distributions. The parameters of these mixture distributions as well as the mixing weights are modelled using Neural Networks.

Deep Recurrent Survival Machines

Deep Recurrent Survival Machines (DRSM) builds on the original DSM model and allows for learning of representations of the input covariates using Recurrent Neural Networks like LSTMs, GRUs. Deep Recurrent Survival Machines is a natural fit to model problems where there are time dependendent covariates. Examples include situations where we are working with streaming data like vital signs, degradation monitoring signals in predictive maintainance. DRSM allows the learnt representations at each time step to involve historical context from previous time steps. DRSM implementation in dsm is carried out through an easy to use API, DeepRecurrentSurvivalMachines that accepts lists of data streams and corresponding failure times. The module automatically takes care of appropriate batching and padding of variable length sequences.

Deep Convolutional Survival Machines

Predictive maintenance and medical imaging sometimes requires to work with image streams. Deep Convolutional Survival Machines extends DSM and DRSM to learn representations of the input image data using convolutional layers. If working with streaming data, the learnt representations are then passed through an LSTM to model temporal dependencies before determining the underlying survival distributions.

Example Usage

>>> from dsm import DeepSurvivalMachines
>>> from dsm import datasets
>>> # load the SUPPORT dataset.
>>> x, t, e = datasets.load_dataset('SUPPORT')
>>> # instantiate a DeepSurvivalMachines model.
>>> model = DeepSurvivalMachines()
>>> # fit the model to the dataset.
>>> model.fit(x, t, e)
>>> # estimate the predicted risks at the time
>>> model.predict_risk(x, 10)

Installation

foo@bar:~$ git clone https://github.com/autonlab/DeepSurvivalMachines.git
foo@bar:~$ cd DeepSurvivalMachines
foo@bar:~$ pip install -r requirements.txt

Examples

1. Deep Survival Machines on the SUPPORT Dataset

References

Please cite the following papers if you are using the dsm package.

[1] Deep Survival Machines: Fully Parametric Survival Regression and Representation Learning for Censored Data with Competing Risks. IEEE Journal of Biomedical \& Health Informatics (2021)

  @article{nagpal2020deep,
  title={Deep Survival Machines: Fully Parametric Survival Regression and\
  Representation Learning for Censored Data with Competing Risks},
  author={Nagpal, Chirag and Li, Xinyu and Dubrawski, Artur},
  journal={IEEE Journal of Biomedical and Health Informatics},
  year={2021}
  }

[2] Recurrent Deep Survival Machines: Deep Parametric Time-to-Event Regression with Time-Varying Covariates. AAAI Spring Symposium (2021)

  @article{nagpal2021rdsm,
  title={Deep Parametric Time-to-Event Regression with Time-Varying Covariates},
  author={Nagpal, Chirag and Jeanselme, Vincent and Dubrawski, Artur},
  journal={AAAI Spring Symposium on Survival Analysis},
  year={2021}
  }

Compatibility

dsm requires python 3.5+ and pytorch 1.1+.

To evaluate performance using standard metrics dsm requires scikit-survival.

Contributing

dsm is on GitHub. Bug reports and pull requests are welcome.

License

MIT License

Copyright (c) 2020 Carnegie Mellon University, Auton Lab

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Sub-modules

dsm.datasets

Utility functions to load standard datasets to train and evaluate the Deep Survival Machines models.

dsm.dsm_api

This module is a wrapper around torch implementations and provides a convenient API to train Deep Survival Machines.

dsm.dsm_torch

Torch model definitons for the Deep Survival Machines model

This includes definitons for the Torch Deep Survival Machines module. The main interface is the DeepSurvivalMachines class which inherits from torch.nn.Module.

Note: NOT DESIGNED TO BE CALLED DIRECTLY!!!

dsm.losses

Loss function definitions for the Deep Survival Machines model

In this module we define the various losses for the censored and uncensored instances of data corresponding to Weibull and LogNormal distributions. These losses are optimized when training DSM.

TODO

Use torch.distributions

Warning

NOT DESIGNED TO BE CALLED DIRECTLY!!!

dsm.utilities

Utility functions to train the Deep Survival Machines models