This is the codebase for the L46 project Personalized Federated Learning in Real-world Healthcare Applications across Different Frameworks.
We introduce how to set up the environment and download the datasets here, and details related to the two frameworks will be introduced in the respective folder. The documentation of planning and decision making through the project is in ./schedule.md.
Set up the environment for using the dataset:
cd FLamby
make install
conda activate flamby
If the make command does not work ((like in Windows, or as in acs-gpu, ended up with the following error)
/bin/sh: 2: /opt/conda/envs/flamby/bin/pip: not found
make: *** [Makefile:8: install] Error 127Try this instead:
cd FLamby
conda env create -f environment.yml
conda activate flamby
pip install -e .[all_extra]
In HPC we also need to run the followinig before conda activate
conda init bash
source ~/.bash_profile
Another useful command to debug with HPC environment when encountering error with PyTorch:
pip3 install torch==1.13.0+cu116 torchvision==0.14.0+cu116 -f https://download.pytorch.org/whl/torch_stable.html
The command line to download the datasets:
Fed-Heart-Disease
cd FLamby/flamby/datasets/fed_heart_disease/dataset_creation_scripts
python download.py --output-folder ./heart_disease_dataset
Fed-IXI
cd FLamby/flamby/datasets/fed_ixi/dataset_creation_scripts
python download.py -o IXI-Dataset
Fed-ISIC2019
cd FLamby/flamby/datasets/fed_isic2019/dataset_creation_scripts
python download_isic.py --output-folder ./isic2019_dataset
python resize_images.py
The main function for comparing the strategies using MsPFL framework is in MsPFL/my_main.py. Please refer to ./MsPFLfolder/README.md for help to execute and navigate the code in detail. The experimental results of the five strategies (four in the report + metaFed) is also included.
We further provide specific instructions in ./Flowerfolder/README.md. Experimental results are shown in the table within the folder.