Federated-CPS

About this Repo

This repository is a base template for Federated Learning build using Flower. Current setup uses LSTM and k-Means with Federated Average for the prediction of geo-altitude for Aviation Dataset.

Pre-requisite

• Docker Engine with (WSL Enabled on Windows)
• Any Code Editor

Folder Structure

.
├── client-1..N             # Code containing the ML/DL algo. in FL setting (acts as FL clients)
├── server                  # Code containing the ML/DL algo. in FL setting (acts as FL server)
├── monitoring              # Code containing the Grafana, Prometheus configuration
├── single-client           # Code containing the ML/DL algo. without FL setting
├── dataset                 # Contains the raw and processed dataset of the aircraft
├── misc                    
└── README.md

Steps to Run

Below are the setup that needs to be done in sequence.

1. Configuration

• Create a SUBNET with a name inside docker using the command: (One-Time Only)
  docker network create --subnet 192.0.0.100/24 <name-of-subnet>

2. Setting Docker Image

• To create the image of federated server
  cd server && docker build -t <name-of-image> .
  (i.e) cd server && docker build -t fl-server .

• To create the image of federated clients
  cd client-1 && docker build -t <name-of-image> .
  (i.e) cd client-1 && docker build -t fl-client-1 .

Note: Perform the same for N-Clients.

3. Running the Docker Image as Docker Container

• [Optional] To run the monitoring tools
  To Start Service: cd monitoring && docker-compose up -d
  To Stop Service: cd monitoring && docker-compose down

• To run the federated server
  docker run -d --network=<name-of-subnet> --ip=<any-IP-within-subnet> --name=<custom-name> <name-of-image>
  (i.e) docker run -d --network=custom-net --ip=192.0.0.101 --name=fl-server fl-server --num_rounds=<any-number>

• To run the federated client-1
  docker run -d --network=<name-of-subnet> --ip=<any-IP-within-subnet-for-client> --name=<custom-name> <name-of-image> --ip=<give-IP-of-FedServer --epochs=<number-of-iter> --folder=<dataset-folder>
  (i.e) docker run -d --network=custom-net --ip=192.0.0.102 --name=fl-client-2 fl-client-2 --ip=192.0.0.101 --epochs=10 --folder=batch

Note: Perform the same for N-Clients. Donot change the PORT as it is pre-configured inside

Arguments

-d : Run in detached mode (in background)