def CW_10way(self,
way,
order,
examples=200,
split=30,
data_len=1024,
shuffle=False,
normalize=True,
label=False,
SNR=None):
"""
1. examples each file <= 119 * 1024
2. if examples>=119, the value of overlap should be True
"""
file_dir = [self.case10[order]]
print('CW_{}way load [{}] loading ……'.format(way, order))
n_way = len(file_dir[0]) # 10 way
assert n_way == way
n_file = len(file_dir) # how many files for each way
num_each_file = examples
num_each_way = num_each_file * n_file
data_size = num_each_file * data_len
data_set = None
for i in range(n_way):
data_ = np.zeros([n_file, num_each_file, data_len])
for j in range(n_file):
data = get_data_csv(file_dir=file_dir[j][i],
num=data_size,
header=-1,
shift_step=200)
data = data.reshape([-1, data_len])
data_[j] = data
data_ = data_.reshape([-1, data_len]) # [num_each_way, 2048]
if SNR is not None:
for k, signal in enumerate(data_):
data_[k] = add_noise(signal, SNR)
if normalize:
data_ = normalization(data_)
if i == 0:
data_set = data_
else:
data_set = np.concatenate((data_set, data_), axis=0)
data_set = data_set.reshape([n_way, num_each_way, data_len,
1])[:, :examples]
if shuffle:
data_set = sample_shuffle(data_set) # 数据少 不建议打乱 不利于训练和测试集有序分开
train_data, test_data = data_set[:, :
split], data_set[:,
split:] # 先shuffle
if label:
label = np.arange(n_way)[:, np.newaxis]
label = np.tile(label, (1, examples)) # [Nc, examples]
train_lab, test_lab = label[:, :split], label[:, split:]
return train_data, train_lab, test_data, test_lab # [Nc,num_each_way,2048,1], [Nc, 50]
else:
return train_data, test_data # [Nc, num_each_way, 2048, 1]
def SA_37way(self,
way,
examples=200,
split=30,
data_len=1024,
shuffle=False,
normalize=True,
label=False):
file_dir = None
if way == 3:
file_dir = self.SA3
elif way == 7:
file_dir = self.SA7
print('SA_{}way loading ……'.format(way))
n_way = len(file_dir[0]) # 3/7 way
assert n_way == way
n_file = len(file_dir) # how many files for each way, 1
num_each_file = examples
num_each_way = num_each_file * n_file
data_size = num_each_file * data_len
data_set = None
for i in range(n_way):
data_ = np.zeros([n_file, num_each_file, data_len])
for j in range(n_file):
data = get_data_csv(file_dir=file_dir[j][i],
num=data_size,
header=-1,
shift_step=200)
data = data.reshape([-1, data_len])
data_[j] = data
data_ = data_.reshape([-1, data_len]) # [num_each_way, 2048]
if normalize:
data_ = normalization(data_)
if i == 0:
data_set = data_
else:
data_set = np.concatenate((data_set, data_), axis=0)
data_set = data_set.reshape([n_way, num_each_way, data_len,
1])[:, :examples]
if shuffle:
data_set = sample_shuffle(data_set) # 数据少 不建议打乱 不利于训练和测试集有序分开
train_data, test_data = data_set[:, :
split], data_set[:,
split:] # 先shuffle
if label:
label = np.arange(n_way)[:, np.newaxis]
label = np.tile(label, (1, examples)) # [Nc, examples]
train_lab, test_lab = label[:, :split], label[:, split:]
return train_data, train_lab, test_data, test_lab # [Nc,num_each_way,1024,1], [Nc, 100]
else:
return train_data, test_data # [Nc, num_each_way, 1024, 1]
def SA_spd(self,
way,
examples=100,
split=30,
data_len=1024,
normalize=True,
label=False,
n_spd=3):
file_dir = [self.SA_speed]
print('SA_{}speed loading ……'.format(n_spd))
n_way = len(file_dir[0][0]) # 3 way
assert n_way == way
n_file = len(file_dir) # how many files for each way, 1
# num_each_file = 5 * 45 if overlap else 45 # 2, 3, 4, 5...
"""
1. examples each file <= 48, suggest: 45
2. if examples>=48, the value of overlap should be True
"""
num_each_spd = examples
data_size = num_each_spd * data_len
train_data = np.zeros([n_way, n_spd, n_file, split, data_len])
test_data = np.zeros(
[n_way, n_spd, n_file, num_each_spd - split, data_len])
for i in range(n_way):
for j in range(n_spd):
for k in range(n_file):
data = get_data_csv(file_dir=file_dir[k][j][i],
num=data_size,
header=-1,
shift_step=200)
data = data.reshape([-1, data_len])[:examples]
# data_[j] = data
if normalize:
data = normalization(data)
train_data[i, j, k] = data[:split]
test_data[i, j, k] = data[split:]
train_data = train_data.reshape([n_way, n_spd, -1, data_len])
test_data = test_data.reshape([n_way, n_spd, -1, data_len])
if label:
label = np.arange(n_way)[:, np.newaxis, np.newaxis]
label = np.tile(label, (1, n_spd, num_each_spd)) # [Nc, examples]
train_lab, test_lab = label[:, :, split], label[:, :, split:]
return train_data, train_lab, test_data, test_lab
# [Nc, n_spd, num_each_way, 2048, 1], [Nc, n_spd, num_each_way]
else:
return train_data, test_data # [Nc, n_spd, num_each_way, 2048, 1]