Skip to main content

Part 1: Launch the FL Experiment

In this first part, we'll set up the Aggregator API that coordinates the federated learning process. The aggregator will be responsible for:

  • Initializing the global model
  • Coordinating training rounds
  • Aggregating model updates
  • Evaluating model performance

Step 1: Install the FL Aggregator API​

Download or clone the API code into your SyftBox API directory

git clone https://github.com/openmined/fl_aggregator.git $HOME/SyftBox/apis/fl_aggregator

Step 2: Configure Your FL Project​

  1. Create a new launch configuration directory:
cp -R $HOME/SyftBox/apis/fl_aggregator/samples/launch_config my_mnist_fl_project
  1. Edit the fl_config.json file in your new directory. Here's a template:
{
  "project_name": "[YOUR_CREATIVE_NAME]_MNIST_FL",
  "aggregator": "your.email@domain.com",
  "participants": [
    "your.email@domain.com",
    "valerio@openmined.org",
    "ionesio@openmined.org"
  ],
  "model_arch": "model.py",
  "model_class_name": "FLModel",
  "model_weight": "global_model_weight.pt",
  "test_dataset": "mnist_test_dataset.pt",
  "rounds": 3,
  "epoch": 10,
  "learning_rate": 0.1
}
note

📝 Please make sure to add your email as both the aggregator and as a participant in the list. In this tutorial, you'll play both roles: coordinating the federated learning experiment as an aggregator and training the model locally as a client.

info

💡 Feel free to add valerio@openmined.org and ionesio@openmined.org as participants in your experiment, as shown in the example! We'd love to collaborate and help you learn. Just remember to send us a quick message on Slack when you've added us, so we can accept your training requests and join your federated learning journey.

tip

🤓 Embrace Your Inner TechnoMage: When choosing your project name, let your inner nerd shine! Whether you're a Trekkie, a Tolkien enthusiast, or a cyberpunk aficionado, pick something that reflects your favorite sci-fi or fantasy universe. From NeuroCounter_MNIST_FL to DigitsSorcery_MNIST_FL, or even ByteMancer_MNIST_FL - make it memorable, make it yours! After all, who said federated learning can't be a bit magical? ✨

Step 3: Launch Your FL Project​

Copy all your configuration files to the launch directory:

cp my_mnist_fl_project/* $HOME/SyftBox/datasites/your.datasite/api_data/fl_aggregator/launch

The API will automatically create all the necessary project directories, set up participant folders, initialize the metrics dashboard, and start the federated learning process.

Step 4: Set Up Test Data​

  1. Create a new fl_aggregator folder into your SyftBox private data directory:
mkdir -p $HOME/SyftBox/private/fl_aggregator
  1. Copy the test MNIST data from the FL Aggregator API samples/test_data directory:
cp $HOME/SyftBox/apis/fl_aggregator/samples/test_data/mnist_test_dataset.pt $HOME/SyftBox/private/fl_aggregator/

Step 5: Monitor Your Experiment Setup​

  1. Navigate to your project's metrics dashboard:
https://syftbox.openmined.org/datasites/your.datasite/fl/[YOUR_CREATIVE_NAME]_MNIST_FL/
  1. You should see your project initialization and dashboard with:
    • Project status and configuration
    • Participant list
    • Training round counters
    • Model evaluation metrics (will populate during training)

Keep this dashboard open in your browser - you'll use it to monitor the training process and verify that all participants have joined your experiment.

🎉 Congratulations! You've successfully set up your federated learning experiment! Your SyftBox Aggregator API is now ready to coordinate the training process across all participants. Continue to Part 2 to learn how to set up the Client API for federated model training.