Federated Learning Workshop with Flower and PyTorch

This guide walks through setting up a federated learning environment using Flower and PyTorch. It covers creating a virtual environment, generating federated datasets, and running the server and client instances.

---

01. Create a Virtual Environment

To begin, create a virtual environment using Conda:

conda create -n fed-env python=3.11.9

After that, activate the environment:

conda activate fed-env

Finally, install the necessary dependencies:

pip install -r requirements.txt

---

02. Create Federated MNIST Dataset for Three Clients

Generate and save the federated MNIST dataset for three clients by running the following command:

First create a save directory

mkdir save

python create_fedatasets.py --root_dir=./data --save_dir=./save

The output will indicate the progress of downloading and saving the datasets:

INFO:root:The downloading is completed
INFO:root:saving federated mnist for client 1 ...
INFO:root:saving federated mnist for client 2 ...
INFO:root:saving federated mnist for client 3 ...
INFO:root:saving test dataset ...
INFO:root:finished

Once completed, the datasets will be stored as follows:

./save
├── mnist-client-1.pt
├── mnist-client-2.pt
├── mnist-client-3.pt
└── mnist-global-test-set.pt

Note: The MNIST dataset will be downloaded the first time you run the script.

---

03. Run the Server and Clients

To initiate the federated learning process, start the server:

python server.py --data_dir="./save" --rounds=6

Next, run the client instances. Start by running client 1:

python client.py --client_id=1 --data_dir=./save

Then, run client 2:

python client.py --client_id=2 --data_dir=./save

Finally, run client 3:

python client.py --client_id=3 --data_dir=./save

---

By following these steps, you can successfully set up a federated learning environment using Flower and PyTorch. Each client trains on its local MNIST data, and the server coordinates the aggregation of their models.