This repository contains the code to run and reproduce the experiments of federated AdaBoost for survival analysis.
It is highly recommended to create a virtual environment prior to installing OpenFL-x (with both venv or conda) to avoid compatibility issues with existing Python software already installed on the system. Furthermore, since OpenFL-x is an extended version of OpenFL, it integrates all the features of the base framework plus the federated bagging and boosting functionalities: it is then suggested to not install both software in the same virtual environment since this can lead to issues in the working of both software.
You can get the last version of OpenFL-x from pypi:
pip install openfl-x
or, alternatively, you can clone this repository and run pip install:
git clone https://github.com/alpha-unito/OpenFL-extended.git
cd OpenFL-extended
pip install .
If this procedure completes successfully, you now have access to all the base features of OpenFL, plus the distributed bagging and boosting functionalities.
Replace the installed openfl/component and openfl/interface directories with the ones provided in this repository to extend the original OpenFL-x framework for working on regression tasks such as survival analysis.
Enjoy!
Run the examples available in the Survival Analysis example folder.
The quickest way to test OpenFL-x is to execute the examples available in the openfl-tutorials/boosting-examples folder. There are many of them available to be run out of the box, each employing a different dataset and a different number of participants in the federation:
The user can customise each example by changing the data distribution across the envoys, the number of envoys itself, the ML model used as weak learner, and the federation's aggregation algorithm. More information on how to run and personalise the examples are available in the openfl-tutorials/boosting-examples folder.
This work is currently under review at the European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. The paper's citation and link will be provided as soon as they become available.
This software is developed and maintained by Oussama Harrak.