def __init__(self, trans_var):
super().__init__(trans_var=trans_var)
self.model = None
self.aggregator = secure_mean_aggregator.Server(self.transfer_variable.secure_aggregator_trans_var)
self.loss_scatter = loss_scatter.Server(self.transfer_variable.loss_scatter_trans_var)
self.has_converged = has_converged.Server(self.transfer_variable.has_converged_trans_var)
self._summary = dict(loss_history=[], is_converged=False)
def server_init_model(self, param):
self.aggregate_iteration_num = 0
self.aggregator = secure_mean_aggregator.Server(
self.transfer_variable.secure_aggregator_trans_var)
self.loss_scatter = loss_scatter.Server(
self.transfer_variable.loss_scatter_trans_var)
self.has_converged = has_converged.Server(
self.transfer_variable.has_converged_trans_var)
self._summary = dict(loss_history=[], is_converged=False)
self.param = param
self.enable_secure_aggregate = param.secure_aggregate
self.max_aggregate_iteration_num = param.max_iter
early_stop = self.model_param.early_stop
self.converge_func = converge_func_factory(early_stop.converge_func,
early_stop.eps).is_converge
self.loss_consumed = early_stop.converge_func != "weight_diff"
def secure_aggregator_call(job_id, role, ind, *args):
if role == consts.ARBITER:
agg = secure_mean_aggregator.Server()
model = agg.weighted_mean_model()
agg.send_aggregated_model(model)
else:
agg = secure_mean_aggregator.Client()
# disorder dit
order = list(range(5))
np.random.seed(random.SystemRandom().randint(1, 100))
np.random.shuffle(order)
raw = {k: np.random.rand(10, 10) for k in order}
w = OrderDictWeights(copy.deepcopy(raw))
d = random.random()
agg.send_weighted_model(w, weight=d)
aggregated = agg.get_aggregated_model()
return aggregated, raw, d
def __init__(self):
self.aggregator = secure_mean_aggregator.Server(
enable_secure_aggregate=True)
self.suffix = tuple()