def read_data_pad_zeros(data_path, split="Train", preprocess=True, fs=500, newFs=256, winSecond=10, winNum=10, n_index=0,
pre_type="sym"):
""" Read data """
# Fixed params
# n_index = 0
n_class = 9
winSize = winSecond * fs
new_winSize = 23296 # winSecond * newFs
# Paths
path_signals = data_path # os.path.join(data_path, split)
# Read time-series data
channel_files = os.listdir(path_signals)
channel_files.sort()
n_channels = 12 # len(channel_files)
X = np.zeros((len(channel_files), new_winSize, n_channels)).astype('float32')
channel_name = ['V6', 'aVF', 'I', 'V4', 'V2', 'aVL', 'V1', 'II', 'aVR', 'V3', 'III', 'V5']
for i_ch, fil_ch in enumerate(channel_files[:]): # tqdm
if i_ch % 1000 == 0:
print(i_ch)
ecg = sio.loadmat(os.path.join(path_signals, fil_ch))
ecg_length = ecg["I"].shape[1]
if ecg_length > 45500:
ecg_length = 45500
ecg_channels = np.zeros((new_winSize, n_channels)).astype('float32')
for i_n, ch_name in enumerate(channel_name):
ecg_data = signal.resample(ecg[ch_name][:, :ecg_length].T,
int(ecg[ch_name][:, :ecg_length].shape[1] / 500 * newFs)).T
if preprocess:
if pre_type == "sym":
ecg_channels[-ecg_data.shape[1]:, i_n] = ecg_preprocessing(ecg_data, 'sym8', 8, 3,
newFs, removebaseline=False,
normalize=False)[0]
elif pre_type == "db4":
ecg_channels[-ecg_data.shape[1]:, i_n] = wavelet(ecg_data[0], 'db4', 4, 2, 4)
elif pre_type == "db6":
ecg_channels[-ecg_data.shape[1]:, i_n] = wavelet_db6(ecg_data[0])
# ecg_channels[:, i_n] = (ecg_channels[:, i_n]-np.mean(ecg_channels[:, i_n]))/np.std(ecg_channels[:, i_n])
else:
pass
print(" no preprocess !!! ")
X[i_ch, :, :] = ecg_channels
return X
def read_data(data_path, split = "TRAIN",preprocess=False):
""" Read data """
# Fixed params
n_class = 2
n_steps = 2560
# Paths
path_signals = os.path.join(data_path, split)
# Read labels and one-hot encode
label_path = os.path.join(data_path, "reference.txt")
labels = pd.read_csv(label_path, sep='\t',header = None)
# Read time-series data
channel_files = os.listdir(path_signals)
#print(channel_files)
channel_files.sort()
n_channels = 12#len(channel_files)
#posix = len(split) + 5
# Initiate array
list_of_channels = []
X = np.zeros((len(channel_files), n_steps, n_channels))
i_ch = 0
for i_ch,fil_ch in enumerate(channel_files):
#channel_name = fil_ch[:-posix]
#dat_ = pd.read_csv(os.path.join(path_signals,fil_ch), delim_whitespace = True, header = None)
ecg = sio.loadmat(os.path.join(path_signals,fil_ch))
ecg['data'] = signal.resample(ecg['data'].T,2560)
if preprocess:
data = ecg_preprocessing(ecg['data'].T, 'sym8', 8, 3, 256)
X[i_ch,:,:] = data.T#ecg['data']
else:
X[i_ch,:,:] = ecg['data']
# Record names
#list_of_channels.append(channel_name)
# iterate
#i_ch += 1
# Return
return X, labels, list_of_channels
def read_data_seg(data_path,
preprocess=True,
fs=500,
newFs=256,
winSecond=10,
winNum=10,
n_index=0,
pre_type="sym"):
""" Read data """
# Fixed params
n_class = 9
winSize = winSecond * fs
new_winSize = winSecond * newFs
# Paths
path_signals = data_path # os.path.join(data_path, split)
# Read time-series data
channel_files = os.listdir(path_signals)
channel_files.sort()
n_channels = 12
X = np.zeros(
(len(channel_files), new_winSize, n_channels)).astype('float32')
channel_name = [
'V6', 'aVF', 'I', 'V4', 'V2', 'aVL', 'V1', 'II', 'aVR', 'V3', 'III',
'V5'
]
for i_ch, fil_ch in enumerate(channel_files[:]): # tqdm
if i_ch % 1000 == 0:
pass #print(i_ch)
ecg = sio.loadmat(os.path.join(path_signals, fil_ch))
ecg_length = ecg["I"].shape[1]
if ecg_length > fs * winNum * winSecond:
print(" too long !!!", ecg_length)
ecg_length = fs * winNum * winSecond
if ecg_length < 4500:
print(" too short !!!", ecg_length)
break
slide_steps = int((ecg_length - winSize) / winSecond)
if ecg_length <= 4500:
slide_steps = 0
ecg_channels = np.zeros((new_winSize, n_channels)).astype('float32')
for i_n, ch_name in enumerate(channel_name):
ecg_channels[:, i_n] = signal.resample(
ecg[ch_name][:, n_index * slide_steps:n_index * slide_steps +
winSize].T, new_winSize).T
if preprocess:
if pre_type == "sym":
ecg_channels[:, i_n] = ecg_preprocessing(
ecg_channels[:, i_n].reshape(1, new_winSize),
'sym8',
8,
3,
newFs,
removebaseline=False,
normalize=False)[0]
elif pre_type == "db4":
ecg_channels[:, i_n] = wavelet(ecg_channels[:, i_n], 'db4',
4, 2, 4)
elif pre_type == "db6":
ecg_channels[:, i_n] = wavelet_db6(ecg_channels[:, i_n])
# ecg_channels[:, i_n] = (ecg_channels[:, i_n]-np.mean(ecg_channels[:, i_n]))/np.std(ecg_channels[:, i_n])
else:
pass
print(" no preprocess !!! ")
X[i_ch, :, :] = ecg_channels
return X
def read_data_seg(data_path, split="Train", preprocess=False, fs=500, newFs=256, winSecond=10, winNum=10, n_index=0,pre_type="sym"):
""" Read data """
# Fixed params
# n_index = 0
n_class = 9
winSize = winSecond * fs
new_winSize = winSecond * newFs
# Paths
path_signals = os.path.join(data_path, split)
# Read labels and one-hot encode
# label_path = os.path.join(data_path, "reference.txt")
# labels = pd.read_csv(label_path, sep='\t',header = None)
# labels = pd.read_csv("reference.csv")
# Read time-series data
channel_files = os.listdir(path_signals)
# print(channel_files)
channel_files.sort()
n_channels = 12 # len(channel_files)
# posix = len(split) + 5
# Initiate array
list_of_channels = []
X = np.zeros((len(channel_files), new_winSize, n_channels)).astype('float32')
i_ch = 0
channel_name = ['V6', 'aVF', 'I', 'V4', 'V2', 'aVL', 'V1', 'II', 'aVR', 'V3', 'III', 'V5']
channel_mid_name = ['II', 'aVR', 'V2', 'V5']
channel_post_name = ['III', 'aVF', 'V3', 'V6']
for i_ch, fil_ch in enumerate(channel_files[:]): # tqdm
if i_ch % 1000 == 0:
print(i_ch)
ecg = sio.loadmat(os.path.join(path_signals, fil_ch))
ecg_length = ecg["I"].shape[1]
if ecg_length > fs * winNum * winSecond:
print(" too long !!!", ecg_length)
ecg_length = fs * winNum * winSecond
if ecg_length < 4500:
print(" too short !!!", ecg_length)
break
slide_steps = int((ecg_length - winSize) / winSecond)
if ecg_length <= 4500:
slide_steps = 0
ecg_channels = np.zeros((new_winSize, n_channels)).astype('float32')
for i_n, ch_name in enumerate(channel_name):
ecg_channels[:, i_n] = signal.resample(ecg[ch_name]
[:, n_index * slide_steps:n_index * slide_steps + winSize].T
, new_winSize).T
if preprocess:
if pre_type == "sym":
ecg_channels[:, i_n] = ecg_preprocessing(ecg_channels[:, i_n].reshape(1, new_winSize), 'sym8', 8, 3,
newFs, removebaseline=False, normalize=False)[0]
elif pre_type == "db4":
ecg_channels[:, i_n] = wavelet(ecg_channels[:, i_n], 'db4', 4, 2, 4)
elif pre_type == "db6":
ecg_channels[:, i_n] = wavelet_db6(ecg_channels[:, i_n])
else:
pass
print(" no preprocess !!! ")
X[i_ch, :, :] = ecg_channels
return X
