server module

class server.Server(model)

Bases: object

get_clients_test_info(clients=None)

Returns the ids, hierarchies and num_test_samples for the given clients.

Returns info about self.selected_clients if clients=None;

Parameters:clients – list of Client objects.

select_clients(possible_clients, num_clients=20)

Selects num_clients clients randomly from possible_clients.

Note that within function, num_clients is set to

min(num_clients, len(possible_clients)).

Parameters:

• possible_clients – Clients from which the server can select.
• num_clients – Number of clients to select; default 20

Returns:

list of (num_train_samples, num_test_samples)

test_model(clients_to_test=None)

Tests self.model on given clients.

Tests model on self.selected_clients if clients_to_test=None.

Parameters:clients_to_test – list of Client objects.

train_model(num_epochs=1, batch_size=10, minibatch=None, clients=None)

Trains self.model on given clients.

Trains model on self.selected_clients if clients=None; each client’s data is trained with the given number of epochs and batches.

Parameters:

• clients – list of Client objects.
• num_epochs – Number of epochs to train.
• batch_size – Size of training batches.
• minibatch – fraction of client’s data to apply minibatch sgd, None to use FedAvg

Returns:

number of bytes written by each client to server

dictionary with client ids as keys and integer values.

client computations: number of FLOPs computed by each client

dictionary with client ids as keys and integer values.

bytes_read: number of bytes read by each client from server

dictionary with client ids as keys and integer values.

Return type:

bytes_written

update_model()