def assign_data():
''' Assign datasets to corresponding devices
'''
# load datasets for two pis
X_train_1, X_train_2, y_train_1, y_train_2 = read_data()
print('_' * 30 + ' check the consistency ' + '_' * 30)
print('len(X_1)=:{}, len(y_1)=:{}'.format(len(X_train_1), len(y_train_1)))
print('len(X_2)=:{}, len(y_2)=:{}'.format(len(X_train_2), len(y_train_2)))
print('-' * 70)
# send data to pis [X_train_1, y_train_1] is for pi1, [X_train_2, y_train_2] is for pi2
data_1 = np.asarray(np.hstack((X_train_1, y_train_1.reshape(-1, 1))))
data_2 = np.asarray(np.hstack((X_train_1, y_train_1.reshape(-1, 1))))
send("172.24.6.253", "data_01", 12345, data_1[:100, :])
send("172.24.6.253", "data_02", 12345, data_2[:100, :])
def main():
# id of this device
ID = "01"
# define local model
optimizer = tf.train.GradientDescentOptimizer(0.1)
model = Model(num_classes=10, optimizer=optimizer, regu_param=1e-3)
# num of communication round
num_rounds = 100
# num of local iterations
num_epochs = 10
# receive dataset from server
topic = "data_" + ID
dataset = receive("172.24.6.253",
topic=topic,
12345,
self_name=None,
time1=None,
count=1)
for round in range(num_rounds):
# receive model parameter from server
model_params = receive("172.24.6.253",
topic="global_model",
12345,
self_name=None,
time1=None,
count=1)
# check if received global model
if len(model_params) == 0:
print('=' * 60)
print('[INFO] DID NOT RECEIVE GLOBAL MODEL !')
return 0
# update local model
local_model = model.solve_inner(data=dataset,
num_epochs=num_epochs,
batch_size=10)
# upload local model to server
send("172.24.6.253", "local_model", 12345, local_model)
def main():
# num of communication round
num_round = 100
# num of devices
num_devices = 2
# assign datasets to devices
assign_data()
# initialize model
global_model_init = model_initialier()
#print(global_model_init)
### do not change model datatype here, change inside the send function
#model_change = [model_init[0].tolist(),model_init[1].tolist()]
# boardcast model_init to all devices
send("172.24.6.253", "global_model", 12345, global_model_init)
print('send model successfully')
# continue the training for 100 iterations
for i in range(num_round):
## I don't know the meaning of self_name, pls specify in your code
local_model = receive("172.24.6.253", "local_model", 12345, "host",
0.1, num_devices - 1)
print("local_model good")
# aggregate local models
# assume I receive a list of local models
global_model = aggregate(local_model)
# braodcast globel_model to devices
send("172.24.6.253", "global_model", 12345, global_model)