martingeorgiu/spots_federated

Federated learning of ANNs for image classification of spots

Setup

Download the dataset from here, rename the dictionary to dataset and move it to the root of this project.

Usage

Example command to run local simulation of federated learning using FedAvg:

python server.py --simulated --rounds 50 --strategy fedavg --train_epochs 3

You can also run following command ./bin/run-proper-flower.sh to run a real local federation with 3 clients using FedAvg with 50 rounds. Beware that this command is only working on macOS.

For additional documentation of all features of this example repo, feel free to dive into any of the root python files which all can be individually run. Also for simple docs you can add --help flag for basic explanation of the scripts.

Logging

To browse the logs use command tensorboard --logdir=dir_xy and replace dir_xy accordingly:

• classic-models/ - logs from non federated trainings will be logged there
• federated-models/ - logs from federated trainings without using any simulated clients
• federated-models-simulated/ - logs from federated trainings with the use of simulated clients