def stream_model_updates(experiment_id, model_updates, client, secret,
task_id):
yield globalserver_pb2.ModelUpdate(client=client,
secret=secret,
experiment_id=experiment_id,
task_id=task_id)
yield globalserver_pb2.ModelUpdate(model_update=model_updates)
def stream_model(experiment_id, model, global_weights, client, secret, task_id):
if int(os.getenv('SERVER', 1)):
gradient = get_gradient(get_weights(model), global_weights)
yield globalserver_pb2.ModelUpdate(client=client, secret=secret, experiment_id=experiment_id, task_id=task_id)
yield globalserver_pb2.ModelUpdate(model_update=array_to_bytes(gradient))
else:
yield globalserver_pb2.ModelUpdate(client=client, secret=secret, experiment_id=experiment_id, task_id=task_id)
yield globalserver_pb2.ModelUpdate(model_update=array_to_bytes(global_weights))
def stream_model_P2P(experiment_id, model, client, secret, task_id):
model_dict = dict()
# save trees for visualization of the model
model_dict['trees'] = str(model.get_dump())
# save model object itself, str format. pickle returns bytes format, we transform it to str
model_dict['pickle'] = str(pickle.dumps(model))
yield globalserver_pb2.ModelUpdate(client=client, secret=secret, experiment_id=experiment_id, task_id=task_id)
yield globalserver_pb2.ModelUpdate(model_update=json.dumps(model_dict).encode('utf-8'))
def stream_model_RF(experiment_id, model, client, secret, task_id):
# the RF_train_model filled this attribute with a json-like string containing histogram data
model_update = json.dumps(model.model_update)
yield globalserver_pb2.ModelUpdate(client=client, secret=secret, experiment_id=experiment_id, task_id=task_id)
yield globalserver_pb2.ModelUpdate(model_update=model_update.encode('utf-8'))