task_num = 3 # number of tasks per WD
# Load data
task_size = sio.loadmat('./data/MUMT_data_3x3')['task_size']
gain = sio.loadmat('./data/MUMT_data_3x3')['gain_min']
# generate the train and test data sample index
# data are splitted as 80:20
# training data are randomly sampled with duplication if N > total data size
split_idx = int(.8* len(task_size))
num_test = min(len(task_size) - split_idx, N - int(.8* N)) # training data size
mem = MemoryDNN(net = [WD_num*task_num, 120, 80, WD_num*task_num],
net_num=net_num,
learning_rate = 0.01,
training_interval=10,
batch_size=128,
memory_size=1024
)
start_time=time.time()
gain_his = []
gain_his_ratio = []
knm_idx_his = []
m_li=[]
env = MU.MUMT(3,3,rand_seed=1)
for i in range(N):
if i % (N//100) == 0:
print("----------------------------------------------rate of progress:%0.2f"%(i/N))
if i < N - num_test:
M = 6 # number of frequency blocks
Memory = 1024 # capacity of memory structure
# Load data
PathInput = "D:/DL_MEC/Data_4x6/Channels/" # "D:/DL_MEC/Data_4x6/Input/"
PathMaxCompRate = "D:/DL_MEC/Data_4x6/MaxCompRate/MaxCompRate.csv"
MaxCompRate = np.genfromtxt(PathMaxCompRate, delimiter=',')
split_idx = int(.8 * len(os.listdir(PathInput)))
num_test = min(len(os.listdir(PathInput)) - split_idx,
n - int(.8 * n)) # training data size
mem = MemoryDNN(net=[N * M, 120, 80, N * M],
net_num=net_num,
learning_rate=0.01,
training_interval=10,
batch_size=128,
memory_size=1024)
start_time = time.time()
gain_his = []
gain_his_ratio = []
knm_idx_his = []
for i in range(n):
if i % (n // 100) == 0:
print(
"----------------------------------------------rate of progress:%0.2f"
% (i / n))
if i < n - num_test:
rate = sio.loadmat('./data/data_%d' % N)['output_obj']
# increase h to close to 1 for better training; it is a trick widely adopted in deep learning
channel = channel * 1000000
# generate the train and test data sample index
# data are splitted as 80:20
# training data are randomly sampled with duplication if n > total data size
split_idx = int(.8 * len(channel))
num_test = min(len(channel) - split_idx,
n - int(.8 * n)) # training data size
mem = MemoryDNN(net=[N, 120, 80, N],
learning_rate=0.01,
training_interval=10,
batch_size=128,
memory_size=Memory)
start_time = time.time()
rate_his = []
rate_his_ratio = []
mode_his = []
k_idx_his = []
K_his = []
h = channel[0, :]
# initilize the weights by setting case_id = 0.
weight, rate = alternate_weights(0)
print("WD weights at time frame %d:" % (0), weight)
)['output_obj'] # this rate is only used to plot figures; never used to train DROO. Generate by CD method in
# optimization.py
mode = sio.loadmat('./data/data_%d' % N)['output_mode']
# increase h to close to 1 for better training; it is a trick widely adopted in deep learning
channel = channel * 1000000
# generate the train and test data sample index
# data are split as 80:20
# training data are randomly sampled with duplication if n > total data size
split_idx = int(.9 * len(channel)) # channel size: 30000*10
num_test = len(channel) - split_idx
mem = MemoryDNN(net=[N, 60, 40, N],
learning_rate=0.001,
training_interval=training_interval,
test_interval=test_interval,
batch_size=batch_size,
memory_size=memory_size,
output_graph=False)
start_time = time.time()
print('\n' + '=' * 20 + 'Start training' + '=' * 20)
print(
'User num:{}\nChannel num(Time frames):{:,}\nK:{}\nDecode mode:{}\nMemory size:{}\nDelta(K update interval):{}'
.format(N, n, K, decoder_mode, memory_size, Delta))
rate_his = []
rate_his_ratio = []
mode_his = []
k_idx_his = []
K_his = []
test_rate_his = []