environment for using the dataset:

cd FLamby
make install
conda activate flamby

downloading new datasets:

cd FLamby
cd flamby/datasets/fed_heart_disease/dataset_creation_scripts
python download.py --output-folder ./heart_disease_dataset

Main code for the evaluation is in PFL/my_main.py. The uncertainty evaluation is in PFL/uncertainty.py. Main code for early exit ensembles is in PFL/earlyexit.py.

Experiment code can be run by simply executing our the main python file,

python my_main.py --alg fedavg

with only the strategy to be parsed and other parameters defined in the code.

To switch between datasets, only the import statements and definitions in the beginning need to be changed (see the annotation).

The trained models and training logs of server&client performance in each round are stored in PFL/cks/.

The config of ISIC2019 is imported from the FLamby benchmark. The config of PAMAP2 is defined in har_config.py, config of PhysioNet-2016 is defined in audio_config.py.