How to send two separated data to two different collaborators?

[amin-kaveh]

Hi,

I am new in federated learning and openfl. I have two separated (let's call them private) datasets and I want to send them to each collaborators separately. When I review tutorials, it seems that FederatedDataSet gets all the data together and split them into multiple sets and then FederatedModel function sends the same model and splits of data to each collaborators.

This is the code that have have from tutorials:

fl_data = FederatedDataSet(train_data, train_labels, valid_data, valid_labels, batch_size, num_classes)
fl_model = FederatedModel(build_model, optimizer, loss_fn, data_loader= fl_data)

But suppose I have private data that I want to sent to each collaborators:
data for collaborator 1: train_data_1, train_label_1, valid_data_1, valid_label_1
data for collaborator 2: train_data_2, train_label_2, valid_data_2, valid_label_2

Now, I want to send the model and data to each collaborators separately and then train them. Something like this:

send(build_model, train_data_1, train_label_1, valid_data_1, valid_label_1, to_collaborator_1 )
send(build_model, train_data_2, train_label_2, valid_data_2, valid_label_2, to_collaborator_2 )

And then start training them.
What are those commands and how can I train the model?

I really appreciate any help and insight.

[mansishr]

Hi @amin-kaveh, welcome to OpenFL! In general, federated learning assumes that the collaborators are data owners so they already have a copy of the local/private dataset with them. Do you mean that in your case, the collaborators cannot access data?

[AtyDany]

hi amin
did you find any answer for your question, i have same problem, however chatgpt suggested bellow code which is not working:

Initialize an OpenFL workspace

workspace = fx.workspace()
workspace.add_model(hospital1_model, 'hospital1')
workspace.add_model(hospital2_model, 'hospital2')
workspace.add_model(hospital3_model, 'hospital3')

Run the federated learning plan

workspace.run_plan(federated_learning_plan)

Evaluate the global model on the test set

global_model = workspace.get_model()
test_loss, test_accuracy = global_model.evaluate(X_test, y_test)
print(f'Global Model Test Accuracy: {test_accuracy}')