This app, together with the model_local_training_app
, form a semi-FL app using SyftBox
- Run 3 clients (
a, b, c
) withjust run-client a
and the SyftBox cache serverjust run-server
- 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 tob
orc
'sconfig.json
file - 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 toa
'sconfig.json
file - Copy the
participants.json
intoa
'sapi_data/model_aggregator/launch
folder, e.g..clients/[email protected]/datasites/[email protected]/api_data/model_aggregator/launch/participants.json
. - Copy the test data, e.g.
mnist_dataset.pt
into .clients/[email protected]/private/model_aggregator/
- Moving the MNIST data parts in
b
andc
'sprivate
intoprivate/model_local_training
to train - The training logs can be seen at the
apis/model_aggregator/logs
folder fora
, orapis/model_local_training/logs
forb
andc