def build_SharedNN_LabelClient(arg_dict: dict):
client = SharedNN_LabelClient(arg_dict["channel"], arg_dict["logger"], arg_dict["mpc_paras"],
CSVDataLoader(arg_dict["data_path"] + "train.csv"),
CSVDataLoader(arg_dict["data_path"] + "test.csv"),
get_loss(arg_dict["loss"]),
get_metric(arg_dict["metric"]),
arg_dict["task_path"])
return ClientHandle(client, {"record": lambda: client.test_record}, client.start_train)
def build_SecureXGBoost_LabelClient(arg_dict: dict):
client = SecureXGBoost_LabelClient(
arg_dict["channel"],
arg_dict["logger"],
arg_dict["mpc_paras"],
CSVDataLoader(arg_dict["data_path"] + "train.csv"),
CSVDataLoader(arg_dict["data_path"] + "test.csv"),
)
return ClientHandle(client, {"record": lambda: client.test_record},
client.start_train)
def test_credit_logistic(dims):
train_data = CSVDataLoader("Test/TestDataset/credit_default.csv",
list(range(40000)), dims)
train_data.set_random_seed({"seed": 8964})
test_data = CSVDataLoader("Test/TestDataset/credit_default.csv",
list(range(40000, 50000)), dims)
logistic_model = k.Sequential(k.layers.Dense(1, k.activations.sigmoid))
logistic_model.compile(k.optimizers.SGD(0.1), k.losses.mean_squared_error)
metrics = []
start_time = time.time()
for i in range(100000):
if i % 1000 == 0:
xys = test_data.get_batch(None)
pred_ys = logistic_model.predict_on_batch(xys[:, :-1])
metric = AUC_KS(xys[:, -1:], pred_ys)
print("Train round {}, metric {}".format(i, metric))
metrics.append([i, time.time() - start_time] + metric)
else:
xys = train_data.get_batch(32)
loss = logistic_model.train_on_batch(xys[:, :-1], xys[:, -1:])
return metrics
def test_credit_data_2pc():
# Disable GPU since server do not have any computation other than sigmoid
print("\n======== Test mpc SecureXGBoost protocol with Credit Default TestDataset ============\n")
ip_dict = {
0: "127.0.0.1:19001",
1: "127.0.0.1:19002",
2: "127.0.0.1:19003",
3: "127.0.0.1:19004",
4: "127.0.0.1:19005"
}
channel0 = Peer(0, "[::]:19001", 10, ip_dict, 13, logger=Logger(prefix="Channel0:"))
channel1 = Peer(1, "[::]:19002", 10, ip_dict, 13, logger=Logger(prefix="Channel1:", level=1))
channel2 = Peer(2, "[::]:19003", 10, ip_dict, 13, logger=Logger(prefix="Channel2:"))
channel3 = Peer(3, "[::]:19004", 10, ip_dict, 13, logger=Logger(prefix="Channel3:"))
channel4 = Peer(4, "[::]:19005", 10, ip_dict, 13, logger=Logger(prefix="Channel4:"))
mpc_paras = MPCClientParas([2, 3], 4, 0, 1)
config = {
"learning_rate": 0.1,
"sync_info": {
"seed": 8964
},
"max_iteration": 5,
"max_depth": 2,
"reg_lambda" : 1,
"gamma" : 0.,
"col_sample_ratio" : 0.8,
"row_sample_ratio" : 1.,
"batch_size" : None,
"test_batch_size" : None
}
main_client = MainClient(channel0, Logger(prefix="Main client:"), mpc_paras,
AUC_KS, config)
# triplets_provider = TripletProducer(channel1, Logger(prefix="Triplet provider:"), mpc_paras, [2, 3])
data_client0 = FeatureClient(channel2, Logger(prefix="Data client 0:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400)), list(range(30))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400, 500)), list(range(30))))
data_client1 = FeatureClient(channel3, Logger(prefix="Data client 1:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400)), list(range(30, 72))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400, 500)), list(range(30, 72))))
label_client = LabelClient(channel4, Logger(prefix="Lable client:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400)), list(range(72, 73))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv", list(range(400, 500)),
list(range(72, 73))), AUC_KS, "")
main_client_start_th = threading.Thread(
target=main_client.start_train,
)
data_client0_th = threading.Thread(target=data_client0.start_train)
data_client1_th = threading.Thread(target=data_client1.start_train)
label_client_th = threading.Thread(target=label_client.start_train)
# triplets_provider_th = threading.Thread(target=triplets_provider.start_listening)
# triplets_provider_th.start()
data_client0_th.start()
data_client1_th.start()
label_client_th.start()
time.sleep(1)
main_client_start_th.start()
print("====== Stop the triplet provider, the training should be auto exited =========")
main_client_start_th.join()
data_client0_th.join()
data_client1_th.join()
label_client_th.join()
print("====== MPC SharedNN Test finished =============")
from Client.MPCClient import MPCClientParas, ClientMode
from Client.SharedNN.DataProviders import LabelClient
from Utils.Log import Logger
from Client.Data.DataLoader import CSVDataLoader
ip_dict = {
0: "127.0.0.1:19001",
1: "127.0.0.1:19002",
2: "127.0.0.1:19003",
3: "127.0.0.1:19004",
4: "127.0.0.1:19005"
}
mpc_paras = MPCClientParas([2, 3], 4, 0, 1)
channel = Peer(4,
"[::]:19005",
3,
ip_dict,
13,
logger=Logger(prefix="Channel4:"))
label_client = LabelClient(
channel, Logger(prefix="Lable client:"), mpc_paras, ClientMode.Train,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000)), list(range(72, 73))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000, 50000)), list(range(72, 73))), MSELoss(),
AUC_KS)
label_client.start_train()
from Communication.RPCComm import Peer
from Client.MPCClient import MPCClientParas, ClientMode
from Client.SharedNN.DataProviders import FeatureClient
from Utils.Log import Logger
from Client.Data.DataLoader import CSVDataLoader
ip_dict = {
0: "127.0.0.1:19001",
1: "127.0.0.1:19002",
2: "127.0.0.1:19003",
3: "127.0.0.1:19004",
4: "127.0.0.1:19005"
}
mpc_paras = MPCClientParas([2, 3], 4, 0, 1)
channel = Peer(3,
"[::]:19004",
3,
ip_dict,
13,
logger=Logger(prefix="Channel3:"))
feature_client2 = FeatureClient(
channel, Logger(prefix="Data client 1:"), mpc_paras, ClientMode.Train,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000)), list(range(30, 72))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000, 50000)), list(range(30, 72))))
feature_client2.start_train()
def build_SharedNN_FeatureClient(arg_dict: dict):
client = SharedNN_FeatureClient(arg_dict["channel"], arg_dict["logger"], arg_dict["mpc_paras"],
CSVDataLoader(arg_dict["data_path"] + "train.csv"),
CSVDataLoader(arg_dict["data_path"] + "test.csv"))
return ClientHandle(client, dict(), client.start_train)
def test_credit_data_2pc():
# Disable GPU since server do not have any computation other than sigmoid
print(
"\n======== Test mpc SharedNN protocol with Credit Default TestDataset ============\n"
)
ip_dict = {
0: "127.0.0.1:19001",
1: "127.0.0.1:19002",
2: "127.0.0.1:19003",
3: "127.0.0.1:19004",
4: "127.0.0.1:19005"
}
channel0 = Peer(0,
"[::]:19001",
10,
ip_dict,
13,
logger=Logger(prefix="Channel0:"))
channel1 = Peer(1,
"[::]:19002",
10,
ip_dict,
13,
logger=Logger(prefix="Channel1:", level=1))
channel2 = Peer(2,
"[::]:19003",
10,
ip_dict,
13,
logger=Logger(prefix="Channel2:"))
channel3 = Peer(3,
"[::]:19004",
10,
ip_dict,
13,
logger=Logger(prefix="Channel3:"))
channel4 = Peer(4,
"[::]:19005",
10,
ip_dict,
13,
logger=Logger(prefix="Channel4:"))
mpc_paras = MPCClientParas([2, 3], 4, 0, 1)
main_client = MainClient(channel0,
Logger(prefix="Main client:"),
mpc_paras,
in_dim=64,
out_dim=1,
layers=[1],
batch_size=64,
test_batch_size=10000,
test_per_batches=11,
learning_rate=0.1,
max_iter=33)
triplets_provider = TripletProducer(channel1,
Logger(prefix="Triplet provider:"),
mpc_paras, [2, 3])
data_client0 = FeatureClient(
channel2, Logger(prefix="Data client 0:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000)), list(range(30))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000, 50000)), list(range(30))))
data_client1 = FeatureClient(
channel3, Logger(prefix="Data client 1:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000)), list(range(30, 72))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000, 50000)), list(range(30, 72))))
label_client = LabelClient(
channel4, Logger(prefix="Lable client:"), mpc_paras,
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000)), list(range(72, 73))),
CSVDataLoader("Test/TestDataset/Data/credit_default.csv",
list(range(40000, 50000)), list(range(72, 73))),
MSELoss(), AUC_KS, "")
main_client_start_th = threading.Thread(target=main_client.start_train, )
data_client0_th = threading.Thread(target=data_client0.start_train)
data_client1_th = threading.Thread(target=data_client1.start_train)
label_client_th = threading.Thread(target=label_client.start_train)
triplets_provider_th = threading.Thread(
target=triplets_provider.start_listening)
triplets_provider_th.start()
data_client0_th.start()
data_client1_th.start()
label_client_th.start()
time.sleep(1)
main_client_start_th.start()
print(
"====== Stop the triplet provider, the training should be auto exited ========="
)
main_client_start_th.join()
data_client0_th.join()
data_client1_th.join()
label_client_th.join()
print("====== MPC SharedNN Test finished =============")
np.savetxt("mpc_record.csv",
np.array(label_client.test_record),
delimiter=",")
triplets_provider.stop_listening()