FedProx

This repo is the implementation of the paper Federated optimization in heterogeneous networks

Run this repo:

1. Downloaded datasets and oraganize them as follows:

   data
   ├── clinical
   │   ├── clinical_test.csv
   │   └── tmp
   │       ├── beta0.05
   │       │   ├── clinical_node_0.csv
   │       │   ├── clinical_node_1.csv
   │       │   ├── clinical_node_2.csv
   │       │   ├── clinical_node_3.csv
   │       │   └── clinical_node_4.csv
   │       ├── beta0.5
   │       └── ...
   └── ...
   

2. Edit the configuration file conf.py. Some important arguments are:

   • global_epochs: number of global epochs
   • local_epochs: number of local epochs
   • beta: parameter of Dirichlet distribution
   • mean_batch: number of instances used in computing the average in FedMix
   • lambda: coefficient in loss of FedMix
   • lr, momentum: optimizer
   • num_parties: number of parties
   • ...

3. Start training:

   python main.py
   #or
   python main.py --global_epoch=n --beta=m