Model Aggregator App

This app, together with the model_local_training_app, form a semi-FL app using SyftBox

Running in dev mode

0. Setting up: Run 3 clients (a, b, c) with just run-client a and the SyftBox cache server just run-server
1. Aggregator: Install the local training app on b, c: syftbox app install https://github.com/OpenMined/model_local_training --config_path <path_to_config.json> where <path_to_config.json> points to b or c's config.json file
2. Participants: Install the model aggregator app on a: syftbox app install https://github.com/OpenMined/model_aggregator --config_path <path_to_config.json> where <path_to_config.json> points to a's config.json file
3. Aggregator: Copy the participants.json into a's api_data/model_aggregator/launch folder, e.g. .clients/[email protected]/datasites/[email protected]/api_data/model_aggregator/launch/participants.json.
4. Aggregator: Copy the test data, e.g. mnist_dataset.pt into .clients/[email protected]/private/model_aggregator/
5. Participants: Moving the MNIST data parts in b and c's private into private/model_local_training to train
6. Monitoring: The training logs can be seen at the apis/model_aggregator/logs folder for a, or apis/model_local_training/logs for b and c