def mnist_config():
folders = split_to_folders(10)
return {
'model_name': "KERAS_MNIST",
'dataloader_name': "KERAS_MNIST",
'location': folders[0],
}
def mnist_config():
folders = split_to_folders(10)
return {
'task_type': TaskType.KERAS_MNIST,
'train_folder': folders[0],
'test_folder': "",
}
n_learners = 5
first_server_port = 9995
# make n servers
server_processes = []
for i in range(n_learners):
port = first_server_port + i
server = GRPCServer(mli_factory=ExampleMliFactory(), port=port)
server_process = Process(target=server.run)
print("starting server", i)
server_process.start()
server_processes.append(server_process)
# Before we make the grpc clients, ensure that there's an mnist folder for each client
data_folders = split_to_folders(n_learners,
data_split=[1 / n_learners] * n_learners)
# Now make the corresponding grpc clients
all_learner_models = []
for i in range(n_learners):
port = first_server_port + i
ml_system = ExampleGRPCLearnerClient(f"client {i}", f"127.0.0.1:{port}")
ml_system.start()
dataloader_params = {"location": data_folders[i]}
ml_system.setup_ml(dataset_loader_name=dataloader_tag,
dataset_loader_parameters=json.dumps(dataloader_params),
model_arch_name=model_tag,
model_parameters=json.dumps({}))
all_learner_models.append(ml_system)
# now colearn as usual!