# Uncertainty Estimation in Deep Bayesian Survival Models
# *UPDATE 11/16/23: pip package now available. Use "pip install bnnsurv". Tested with TensorFlow 2.13 and TensorFlow Probability 0.21. See [test file](https://github.com/thecml/UE-BNNSurv/blob/main/tests/test_bnn_surv.py) for how to use.
This repository is the official TensorFlow implementation of [Uncertainty Estimation in Deep Bayesian Survival Models](https://ieeexplore.ieee.org/document/10313466), BHI 2023.
The proposed method is implemented based on [TensorFlow Probability](https://github.com/tensorflow/probability).
<b>Full paper is available on IEEE Xplore: https://ieeexplore.ieee.org/document/10313466</b>
<p align="left"><img src="https://github.com/thecml/UE-BNNSurv/blob/main/img/BNN.png" width="40%" height="40%">
In this work, we introduce the use of Bayesian inference techniques for survival analysis in neural networks that rely on the Cox’s proportional hazard assumption, for which we discuss a new flexible and effective architecture. We implement three architectures: a fully-deterministic neural network that acts as a baseline, a Bayesian model using variational inference and one using Monte-Carlo Dropout.
Experiments show that the Bayesian models improve predictive performance over SOTA neural networks in a test dataset with few samples (WHAS500, 500 samples) and provide comparable performance in two larger ones (SEER and SUPPORT, 4024 and 8873 samples, respectively)
<p align="center"><img src="https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_all_models.png" width="30%" height="30%" /> <img src="https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_grade_mcd.png" width="30%" height="30%" /> <img src="https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_pdf.png" width="31%" height="31%" />
License
--------
To view the license for this work, visit https://github.com/thecml/UE-BNNSurv/blob/main/LICENSE
Requirements
----------------------
To run the models, please refer to [Requirements.txt](https://github.com/thecml/UE-BNNSurv/blob/main/requirements.txt).
Install auton-survival manually from Git:
```
pip install git+https://github.com/autonlab/auton-survival.git
```
Code was tested in virtual environment with `Python 3.8`, `TensorFlow 2.11` and `TensorFlow Probability 0.19`
Training
--------
- Make directories `mkdir results` and `mkdir models`.
- Please refer to `paths.py` to set appropriate paths. By default, results are in `results` and models in `models`
- Network configuration using best hyperparameters are found in `configs/*`
- Run the training code:
```
# SOTA models
python train_sota_models.py
# BNN Models
python train_bnn_models.py
```
Evaluation
--------
- After model training, view the results in the `results` folder.
Visualization
---------
- Run the visualization notebooks:
```
jupyter notebook plot_survival_curves.ipynb
jupyter notebook plot_survival_time.ipynb
```
Citation
--------
```
@inproceedings{lillelund_uncertainty_2023,
author={Lillelund, Christian Marius and Magris, Martin and Pedersen, Christian Fischer},
booktitle={2023 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI)},
title={Uncertainty Estimation in Deep Bayesian Survival Models},
year={2023},
pages={1-4},
doi={10.1109/BHI58575.2023.10313466}
}
```
Raw data
{
"_id": null,
"home_page": "https://github.com/thecml/UE-BNNSurv",
"name": "bnnsurv",
"maintainer": "",
"docs_url": null,
"requires_python": "",
"maintainer_email": "",
"keywords": "Deep Learning,Neural Network,Bayesian Learning,Survival Analysis",
"author": "Christian Marius Lillelund",
"author_email": "chr1000@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/59/ee/2148113cc2906e3bcd1759056ac59d59bf53e700a669c998f5c06dc6d122/bnnsurv-0.1.3.tar.gz",
"platform": null,
"description": "# Uncertainty Estimation in Deep Bayesian Survival Models\n\n# *UPDATE 11/16/23: pip package now available. Use \"pip install bnnsurv\". Tested with TensorFlow 2.13 and TensorFlow Probability 0.21. See [test file](https://github.com/thecml/UE-BNNSurv/blob/main/tests/test_bnn_surv.py) for how to use.\n\nThis repository is the official TensorFlow implementation of [Uncertainty Estimation in Deep Bayesian Survival Models](https://ieeexplore.ieee.org/document/10313466), BHI 2023.\n\nThe proposed method is implemented based on [TensorFlow Probability](https://github.com/tensorflow/probability).\n\n<b>Full paper is available on IEEE Xplore: https://ieeexplore.ieee.org/document/10313466</b>\n\n<p align=\"left\"><img src=\"https://github.com/thecml/UE-BNNSurv/blob/main/img/BNN.png\" width=\"40%\" height=\"40%\">\n\nIn this work, we introduce the use of Bayesian inference techniques for survival analysis in neural networks that rely on the Cox\u00e2\u20ac\u2122s proportional hazard assumption, for which we discuss a new flexible and effective architecture. We implement three architectures: a fully-deterministic neural network that acts as a baseline, a Bayesian model using variational inference and one using Monte-Carlo Dropout.\n\nExperiments show that the Bayesian models improve predictive performance over SOTA neural networks in a test dataset with few samples (WHAS500, 500 samples) and provide comparable performance in two larger ones (SEER and SUPPORT, 4024 and 8873 samples, respectively)\n\n<p align=\"center\"><img src=\"https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_all_models.png\" width=\"30%\" height=\"30%\" /> <img src=\"https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_grade_mcd.png\" width=\"30%\" height=\"30%\" /> <img src=\"https://github.com/thecml/UE-BNNSurv/blob/main/img/seer_surv_pdf.png\" width=\"31%\" height=\"31%\" />\n\n\nLicense\n--------\nTo view the license for this work, visit https://github.com/thecml/UE-BNNSurv/blob/main/LICENSE\n\n\nRequirements\n----------------------\nTo run the models, please refer to [Requirements.txt](https://github.com/thecml/UE-BNNSurv/blob/main/requirements.txt).\n\nInstall auton-survival manually from Git:\n```\npip install git+https://github.com/autonlab/auton-survival.git\n```\nCode was tested in virtual environment with `Python 3.8`, `TensorFlow 2.11` and `TensorFlow Probability 0.19`\n\n\nTraining\n--------\n- Make directories `mkdir results` and `mkdir models`.\n\n- Please refer to `paths.py` to set appropriate paths. By default, results are in `results` and models in `models`\n\n- Network configuration using best hyperparameters are found in `configs/*`\n\n- Run the training code:\n\n```\n# SOTA models\npython train_sota_models.py\n\n# BNN Models\npython train_bnn_models.py\n```\n\n\nEvaluation\n--------\n- After model training, view the results in the `results` folder.\n\n\nVisualization\n---------\n- Run the visualization notebooks:\n```\njupyter notebook plot_survival_curves.ipynb\njupyter notebook plot_survival_time.ipynb\n```\n\n\nCitation\n--------\n```\n@inproceedings{lillelund_uncertainty_2023,\n author={Lillelund, Christian Marius and Magris, Martin and Pedersen, Christian Fischer},\n booktitle={2023 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI)}, \n title={Uncertainty Estimation in Deep Bayesian Survival Models}, \n year={2023},\n pages={1-4},\n doi={10.1109/BHI58575.2023.10313466}\n}\n```\n\n\n",
"bugtrack_url": null,
"license": "",
"summary": "TensorFlow 2.x Bayesian Neural Network for Survival Analysis",
"version": "0.1.3",
"project_urls": {
"Homepage": "https://github.com/thecml/UE-BNNSurv"
},
"split_keywords": [
"deep learning",
"neural network",
"bayesian learning",
"survival analysis"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "bec66ba1edf07f3a1497e7ff452f21d2af0171caa7b7e706480fe173f91abe49",
"md5": "ae483b4c7cf7c61a1e9f6819b875d7aa",
"sha256": "5e3186a0af3b657ebddb7ec99f3f0044d7fc97e576aa2e748dd08f0eb1f4ac68"
},
"downloads": -1,
"filename": "bnnsurv-0.1.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "ae483b4c7cf7c61a1e9f6819b875d7aa",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": null,
"size": 9594,
"upload_time": "2023-11-17T13:33:29",
"upload_time_iso_8601": "2023-11-17T13:33:29.895696Z",
"url": "https://files.pythonhosted.org/packages/be/c6/6ba1edf07f3a1497e7ff452f21d2af0171caa7b7e706480fe173f91abe49/bnnsurv-0.1.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "59ee2148113cc2906e3bcd1759056ac59d59bf53e700a669c998f5c06dc6d122",
"md5": "aeb4767e1eb9059c7351ac53c9f0208f",
"sha256": "822843d59b73743d9d97865bc2782bac2092dac37ba85f787ec8175db6045a68"
},
"downloads": -1,
"filename": "bnnsurv-0.1.3.tar.gz",
"has_sig": false,
"md5_digest": "aeb4767e1eb9059c7351ac53c9f0208f",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 10455,
"upload_time": "2023-11-17T13:33:31",
"upload_time_iso_8601": "2023-11-17T13:33:31.433668Z",
"url": "https://files.pythonhosted.org/packages/59/ee/2148113cc2906e3bcd1759056ac59d59bf53e700a669c998f5c06dc6d122/bnnsurv-0.1.3.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-11-17 13:33:31",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "thecml",
"github_project": "UE-BNNSurv",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [],
"lcname": "bnnsurv"
}
JSON
