tomas-balaz/nvflare_example_docker

Federated Learning with MONAI using NVFLare

Brief Introduction

This repository contains an end-to-end Federated training example based on MONAI trainers and NVFlare.

Inside this folder:

• All Jupiter notebooks are used to build an FL experiment step-by-step.
• demo_figs is a folder containing all referred example figures in the notebooks.
• spleen_example is the experiment config folder. Some of the experiment related hyperparameters are set in spleen_example/config/config_train.json. You may need to modify multi_gpu and some other parameters. Please check the docstrings in spleen_example/custom/train_configer.py for more details.
• build_docker_provision.sh is the script to build the docker image and do provision.
• docker_files is a folder containing all files to build the docker image.
• expr_files is a folder containing all files to be used for the experiment.

Inside expr_files:

project.yml is the project yml file to describe the FL project, it defines project name, participants, server name and othe settings. You can keep the default settings, but may need to change the cn name to the server name as:

server:
  cn: <your server name >

authz_config.json is the authorization configuration json file, it defines groups, roles and rights for all users, organizations and sites. If you modified project.yml, please change this file to keep the consistency. prerpare_expr_files.sh is the script to do provision and (optional) download the Decathlon Spleen dataset for this experiment.

Provision Package Preparation

We need to build a docker image for the experiment. It will be based on MONAI's latest docker image in Docker Hub as well as nvflare library in PyPI.

Please ensure that you have installed Docker (https://docs.docker.com/engine/install/).

Please run bash build_docker_provision.sh

Build Experiment

Please ensure that you have installed JupyterLab (https://jupyterlab.readthedocs.io/en/stable/getting_started/installation.html). The following is an example to install JupyterLab in Python Virtual Environment:

python3 -m venv venv/fl_startup
source venv/fl_startup/bin/activate
pip install --upgrade pip
pip install wheel
pip install jupyterlab

Please run the following command:

jupyter lab --ip 0.0.0.0 --port 8888 --allow-root --no-browser --NotebookApp.token=MONAIFLExample

The link is: http://localhost:8888/?token=MONAIFLExample

Then run 1-Startup.ipynb. You should follow the steps in the notebook, which will guide you through the process of building an FL experiment based on 2 clients and 1 server.