def dataload(Path,
inputsrecord,
targets,
location,
leadnum=12,
downsample=2,
buf_size=100):
target_len = int(72000 / downsample)
samplename = inputsrecord[location]
samplelabel = targets[location]
SEG_buf = np.zeros([1, target_len, leadnum], dtype=np.float32)
SEGs = np.zeros([1, target_len, leadnum], dtype=np.float32)
for i in range(len(samplename)):
sig = np.load(Path + samplename[i])
SEGt = np.float32(utils.sig_process(sig, target_length=target_len))
SEG_buf = np.concatenate((SEG_buf, SEGt))
del SEGt
if SEG_buf.shape[0] >= buf_size:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
SEG_buf = np.zeros([1, target_len, leadnum], dtype=np.float32)
if SEG_buf.shape[0] > 1:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
return SEGs[1:], samplelabel
def datarecord2(input_directory,target_lead,buf_size=100,segnum=24,seg_length=750, full_seg=True, stt=0):
input_files = []
for f in os.listdir(input_directory):
if os.path.isfile(os.path.join(input_directory, f)) and not f.lower().startswith('.') and f.lower().endswith('mat'):
input_files.append(f)
classes=get_classes(input_directory,input_files)
num_files = len(input_files)
datalabel=[]
SEG_buf = np.zeros([1, seg_length, segnum], dtype=np.float32)
SEGs = np.zeros([1, seg_length, segnum], dtype=np.float32)
for i, f in enumerate(input_files):
print(' {}/{}...'.format(i + 1, num_files))
tmp_input_file = os.path.join(input_directory, f)
data, header_data = load_challenge_data(tmp_input_file)
datalabel.append(getdata_class(header_data))
datalead = data[target_lead,:]
# SEGt = np.float32(utils.sig_process(data, target_length=target_len))
SEGt = utils.Stack_Segs_generate2(datalead, seg_num=segnum, seg_length=seg_length, full_seg=full_seg, stt=stt)
del data,datalead
SEG_buf = np.concatenate((SEG_buf, SEGt))
del SEGt
if SEG_buf.shape[0] >= buf_size:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
SEG_buf = np.zeros([1, seg_length, segnum], dtype=np.float32)
if SEG_buf.shape[0] > 1:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
datalabel = np.array(datalabel)
return SEGs[1:], datalabel,len(classes)
def minibatches(Path,
inputsrecord,
targets,
batch_size,
leadnum=12,
downsample=2):
i = 0
target_len = int(72000 / downsample)
while 1: # 要无限循环
labels = []
indices = np.arange(len(inputsrecord))
SEG_buf = np.zeros([1, target_len, leadnum], dtype=np.float32)
for b in range(batch_size):
if i == len(inputsrecord):
i = 0
np.random.shuffle(indices)
samplename = inputsrecord[indices[i]]
samplelabel = targets[i]
labels.append(samplelabel)
i += 1
sig = np.load(Path + samplename)
SEGt = np.float32(utils.sig_process(sig, target_length=target_len))
SEG_buf = np.concatenate((SEG_buf, SEGt))
del SEGt
yield SEG_buf[1:], np.array(labels)
def datarecord(input_directory, downsample, buf_size=100, leadnum=12):
input_files = []
for f in os.listdir(input_directory):
if os.path.isfile(
os.path.join(input_directory, f)
) and not f.lower().startswith('.') and f.lower().endswith('mat'):
input_files.append(f)
classes = get_classes(input_directory, input_files)
num_files = len(input_files)
datalabel = []
target_len = int(72000 / downsample)
SEG_buf = np.zeros([1, target_len, leadnum], dtype=np.float32)
SEGs = np.zeros([1, target_len, leadnum], dtype=np.float32)
for i, f in enumerate(input_files):
print(' {}/{}...'.format(i + 1, num_files))
tmp_input_file = os.path.join(input_directory, f)
data, header_data = load_challenge_data(tmp_input_file)
datalabel.append(getdata_class(header_data))
SEGt = np.float32(utils.sig_process(data, target_length=target_len))
del data
SEG_buf = np.concatenate((SEG_buf, SEGt))
del SEGt
if SEG_buf.shape[0] >= buf_size:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
SEG_buf = np.zeros([1, target_len, leadnum], dtype=np.float32)
if SEG_buf.shape[0] > 1:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
datalabel = np.array(datalabel)
return SEGs[1:], datalabel, len(classes)
def __init__(self, sig, fs=250.0):
assert len(sig.shape) == 1, 'The signal must be 1-dimension.'
assert sig.shape[0] >= fs * 6, 'The signal must >= 6 seconds.'
self.sig = utils.WTfilt_1d(sig)
self.fs = fs
self.rpeaks, = ecg.hamilton_segmenter(signal=self.sig,
sampling_rate=self.fs)
self.rpeaks, = ecg.correct_rpeaks(signal=self.sig,
rpeaks=self.rpeaks,
sampling_rate=self.fs)
self.RR_intervals = np.diff(self.rpeaks)
self.dRR = np.diff(self.RR_intervals)
def datafeatrecord(input_directory,
records,
downsample,
buf_size=100,
leadnum=12,
featurenum=25):
# input_files = []
# for f in os.listdir(input_directory):
# if os.path.isfile(os.path.join(input_directory, f)) and not f.lower().startswith('.') and f.lower().endswith('mat'):
# input_files.append(f)
classes = get_classes(input_directory, records)
num_files = len(records)
datalabel = np.zeros([1, 9])
# label0temp=[]
target_len = int(72000 / downsample)
SEG_buf = np.zeros([1, target_len, leadnum + 1], dtype=np.float32)
SEGs = np.zeros([1, target_len, leadnum + 1], dtype=np.float32)
# feat_buf=np.zeros([1,1,target_len], dtype=np.float32)
featurezero = np.zeros([target_len, 1])
for i, f in enumerate(records):
print(' {}/{}...'.format(i + 1, num_files))
tmp_input_file = os.path.join(input_directory, f)
data, header_data = load_challenge_data(tmp_input_file)
labelonhot, label0 = getdata_class(header_data)
datalabel = np.concatenate((datalabel, labelonhot), axis=0)
# label0temp.append(label0)
features = np.asarray(get_12ECG_features(data, header_data))
featurezero[0:featurenum, 0] = features[0:featurenum]
# feats_reshape = features.reshape(1, -1)
feats_reshape = featurezero.reshape(
[1, featurezero.shape[0], featurezero.shape[1]])
# feat_buf=np.concatenate((feat_buf,feats_reshape))
SEGt = np.float32(utils.sig_process(data, target_length=target_len))
SEGt = np.concatenate((SEGt, feats_reshape), axis=2)
del data
SEG_buf = np.concatenate((SEG_buf, SEGt))
del SEGt
if SEG_buf.shape[0] >= buf_size:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
SEG_buf = np.zeros([1, target_len, leadnum + 1], dtype=np.float32)
if SEG_buf.shape[0] > 1:
SEGs = np.concatenate((SEGs, SEG_buf[1:]))
del SEG_buf
# label0temp = np.array(label0temp)
return SEGs[1:], datalabel[1:]
records_label = np.load(config.REVISED_LABEL) - 1
class_num = len(np.unique(records_label))
train_val_records, _, train_val_labels, test_labels = train_test_split(
records_name,
records_label,
test_size=0.2,
random_state=config.RANDOM_STATE)
train_records, val_records, train_labels, val_labels = train_test_split(
train_val_records,
train_val_labels,
test_size=0.2,
random_state=config.RANDOM_STATE)
_, train_labels = utils.oversample_balance(train_records, train_labels,
config.RANDOM_STATE)
_, val_labels = utils.oversample_balance(val_records, val_labels,
config.RANDOM_STATE)
# 载入之前保存的网络输出概率以及人工特征 -------------------------------------------------------------------------------
pred_nnet_r = np.load(config.MODEL_PATH + 'pred_nnet_r.npy')
pred_nnet_v = np.load(config.MODEL_PATH + 'pred_nnet_v.npy')
pred_nnet_t = np.load(config.MODEL_PATH + 'pred_nnet_t.npy')
man_features_r = np.load(config.MAN_FEATURE_PATH + 'man_features_r.npy')
man_features_v = np.load(config.MAN_FEATURE_PATH + 'man_features_v.npy')
man_features_t = np.load(config.MAN_FEATURE_PATH + 'man_features_t.npy')
pred_r = np.concatenate((pred_nnet_r, man_features_r), axis=1)
pred_v = np.concatenate((pred_nnet_v, man_features_v), axis=1)
pred_t = np.concatenate((pred_nnet_t, man_features_t), axis=1)
train_val_records, test_records, train_val_labels, test_labels = train_test_split(
records_name,
records_label,
test_size=0.2,
random_state=config.RANDOM_STATE)
del test_records, test_labels
train_records, val_records, train_labels, val_labels = train_test_split(
train_val_records,
train_val_labels,
test_size=0.2,
random_state=config.RANDOM_STATE)
# 过采样使训练和验证集样本分布平衡 -------------------------------------------------------------------------------------
train_records, train_labels = utils.oversample_balance(train_records,
train_labels,
config.RANDOM_STATE)
val_records, val_labels = utils.oversample_balance(val_records, val_labels,
config.RANDOM_STATE)
# 取出训练集和测试集病人对应导联信号，并进行切片和z-score标准化 --------------------------------------------------------
print('Fetching data ...-----------------\n')
TARGET_LEAD = 1
train_x = utils.Fetch_Pats_Lbs_sLead(train_records,
Path=config.DATA_PATH,
target_lead=TARGET_LEAD,
seg_num=config.SEG_NUM,
seg_length=config.SEG_LENGTH)
train_y = to_categorical(train_labels, num_classes=class_num)
val_x = utils.Fetch_Pats_Lbs_sLead(val_records,
Path=config.DATA_PATH,