def on_init(self, *args):
print('EA on_init')
FederatedClient.on_init(self, *args)
model_config = args[0]
self.p = model_config["p"]
self.e = model_config["e"]
def synchronize():
global_w = self.request_weights()
local_w = self.local_model.get_weights()
diff = [self.e * (w-gw) for w,gw in zip(local_w, global_w)]
self.local_model.set_weights([w-d for w,d in zip(local_w, diff)])
self.send_diff(diff)
def train():
while True:
if random.random() < self.p:
synchronize()
self.local_model.train_one_round()
threading.Thread(target = train).start()
def start_client():
print("start client")
# c = FederatedClient("127.0.0.1", 5000, datasource.Mnist)
FederatedClient("192.168.199.222", 5000, datasource.Mnist)
replace=False)
attack_mode = [0] * len(selected_writers)
attack_number = int(args.attack_ratio * len(selected_writers))
for attack_idx in range(attack_number):
attack_mode[attack_idx] = args.attack_mode
print("Selected Writers", selected_writers,
len(selected_writers))
print("Attack Modes", attack_mode)
# feiminst数据集使用
# for selected_idx in range(len(selected_writers)):
# clients.append( FederatedClient("127.0.0.1", 5000, datasource.Mnist, gpu,
# attack_mode[selected_idx], writer=writers[ selected_writers[selected_idx] ]) )
for selected_idx in range(len(selected_writers)):
clients.append(FederatedClient(gpu,\
attack_mode[selected_idx], writer=writers[ selected_writers[selected_idx] ], number=selected_writers[selected_idx]) )
init_info = {
# 'model_json': server.global_model.model.to_json(),
'model_json': server.global_model.model,
'model_id': server.model_id,
'min_train_size': 500,
'data_split': (0.6, 0.3, 0.1), # train, test, valid
'epoch_per_round': args.local_epoch,
'batch_size': args.batch_size,
'request_sid': "rid", #request.sid,
}
for client in clients:
client.on_init(init_info)
########
def __init__(self, server_host, server_port, datasource):
# probability to synchronize. Note: here epoch_per_round ~ 1/p
self.p = None
self.e = None # weight for elasiticity term
FederatedClient.__init__(self, server_host, server_port, datasource)
def start_client():
print("start client")
c = FederatedClient("127.0.0.1", 5000, datasource.Mnist)
def start_client(cid, ip, port, thread):
print("start client")
c = FederatedClient(ip, int(port), thread, cid)
def start_client0():
print("start client0")
c = FederatedClient("172.17.0.2", 1111, datasource5.Mnist)
def start_client():
print("start client")
c = FederatedClient(server_host, server_port, datasource.Mnist)
def start_client(value_data_class_train_size):
device_id = value_data_class_train_size['device_id'] - 1
print("start client on {}".format(device_id))
with tf.device('/gpu:{}'.format(device_id)):
c = FederatedClient(value_data_class_train_size, "127.0.0.1", 5000,
datasource.Mnist)
def start_client(fo_name, f_training_name):
#c = FederatedClient("172.31.20.224", 5000, datasource_CIFAR10.Cifar10, fo_name, f_training_name)
c = FederatedClient("172.31.20.224", 5000, datasource.Mnist, fo_name,
f_training_name)