This guide explains how to set up and run a classification federated learning system using Docker. The setup involves multiple clients and a central server, all managed via a docker.sh script. ###Instructions ####1. Setting Up Clients and Server Execute: Use the docker.sh script to start both clients and the server: see _FIles_ Adjust Parameters: Modify the script settings as required for each client and the server. ####2. Script Configuration **MY_DATA_DIR** Purpose: Defines the path to the data directory. Example Configuration: ``` MY_DATA_DIR=/home/Pictures ``` Directory Structure: Organize your data directory as follows: ``` . ├── center01 │ ├── Train_Private │ │ ├── 1 │ │ ├── 12 ... │ │ └── train_private.csv │ └── Train_Public │ ├── 14 ... │ └── train_public.csv ├── center02 ... │ ├── Train_Private │ │ ├── 10 ... │ │ ├── 9 │ │ └── train_private.csv │ └── Train_Public │ ├── 16 ... │ └── train_public.csv ├── center03 │ ├── Train_Private │ │ ├── 1 .. │ │ └── train_private.csv │ └── Train_Public │ ├── 106 ... │ └──train_public.csv ``` **Note:** ! To Train a Model: run the ``` prepare_data_top5.py ``` script where you need to provide ``` --root_dir 'app/data/center03/Train_Private' ``` and ``` --center 'center03' ``` **GPU2USE** Purpose: Specifies the GPU resources to be used. Options: All GPUs: ``` GPU2USE=all ``` Specific GPUs: ``` GPU2USE='"device=0,2"' ``` **NETARG** Purpose: Defines network arguments for Docker containers. Default Setting: ``` NETARG="--net=host" ``` Modification: Change this if additional network settings are needed. **CONTAINER_NAME** Purpose: Sets the name of the Docker container. Configuration: For Clients: Set to the specific client name (e.g., client1, client2, etc.): ``` CONTAINER_NAME=client1 ``` For Server: Set to server: ``` CONTAINER_NAME=server ``` ####2. Running a server 1. Run ./docker.sh 2. Edit config_server.yaml located in ./training/src/server if needed to provide a relevant address to a server 3. Launch the server ``` python ./training/src/server/server_fl.py ``` ####3. Running a client 1. Run ./docker.sh 2. Edit config_server.yaml located in ./training/src/client if needed to provide a relevant address to a server 3. Start a client n (n = 1, 2, 3) ``` python ./training/src/client/clientn.py ``` ####4. Results All results are saved as weights to _output_ folder. For example, the weights models trained on public portion of the dataset can be found here: ``` /training/output_public/models ```