def make_set(df_data, channels, label_map, record_id, window_size=2048):
n_windows = 0
for _, record in tqdm(df_data.iterrows()):
n_windows += record['signal_length'] // window_size
dataX = np.zeros((n_windows, len(channels), window_size))
dataY = np.zeros((n_windows, len(label_map)))
record_list = []
nth_window = 0
for i, (patient, record) in enumerate(tqdm(df_data.iterrows())):
# read the record, get the signal data and transpose it
signal_data = io.rdrecord(os.path.join(
'data', record['name'])).p_signal.transpose()
n_rows = signal_data.shape[-1]
n_windows = n_rows // window_size
dataX[nth_window:nth_window + n_windows] = np.array([
signal_data[:, i * window_size:(i + 1) * window_size]
for i in range(n_windows)
])
dataY[nth_window:nth_window + n_windows][:,
label_map[record.label]] = 1
nth_window += n_windows
if record_id:
record_list += n_windows * [record['name']]
return dataX, dataY, record_list
def get_record(record_names, prefix=''):
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(record_name=os.path.join(data_folder, record_name))
data = np.transpose(np.mean(record.p_signal, axis=1))
records.append(data)
return np.array(records)
def extract_from_wfdb(wfdb_file_path, include_channels, header, **kwargs):
"""
TODO
Returns:
ndarray, shape NxC
"""
from wfdb.io import rdrecord
rec = rdrecord(record_name=os.path.splitext(wfdb_file_path)[0],
channel_names=include_channels)
header["sample_rate"] = float(rec.fs)
return rec.p_signal
def make_set(df_data,
label_map,
record_id,
indices_channels,
window_size=2048,
taux_echant=1):
n_channels = len(indices_channels)
overlapping = 2
n_windows = get_len_set(df_data, overlapping, window_size, taux_echant)
dataX = np.zeros((n_windows, n_channels, window_size))
dataY = np.zeros((n_windows, len(label_map)))
record_list = []
nth_window = 0
for i, (patient, record) in tqdm(enumerate(df_data[:5].iterrows())):
signal_data = io.rdrecord(os.path.join(
data_folder, record['name'])).p_signal.transpose()
data_for_patient, n_rows, n_windows = format_X(signal_data,
taux_echant,
overlapping,
indices_channels,
window_size)
dataX[nth_window:nth_window + n_windows] = data_for_patient
dataY[nth_window:nth_window + n_windows][:,
label_map[record.label]] = 1
last_id = nth_window + n_windows
nth_window += n_windows
if record_id:
record_list += n_windows * [record['name']]
dataX = dataX[:nth_window]
dataY = dataY[:nth_window]
if not record_id:
X_twa, Y_twa = get_twadb_db()
print(X_twa.shape)
print(Y_twa.shape)
print(dataX.shape)
print(dataY.shape)
dataX = np.array(dataX.tolist() + X_twa.tolist())
dataY = np.array(dataY.tolist() + Y_twa.tolist())
print('nth_window', nth_window)
print('last id', last_id)
print('len dataX', len(dataX))
return dataX, dataY, record_list
def get_record(record_names, prefix='../'):
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(
record_name=os.path.join(data_folder, record_name))
label = record.comments[4].split(": ")[-1]
patient = record_name.split('/')[0]
signal_length = record.sig_len
records.append({
'name': record_name,
'label': label,
'patient': patient,
'signal_length': signal_length
})
return pd.DataFrame(records)
def load_wfdb(file_path, load_channels, **kwargs):
"""
File loader function for .dat and .mat WFDB extension files.
Redirects to the wfdb.io.rdrecord function.
Returns:
wfdb.io.record.Record object
"""
from wfdb.io import rdrecord
rec = rdrecord(record_name=os.path.splitext(file_path)[0],
channel_names=load_channels)
chans = rec.sig_name
load_channels = load_channels or chans
if not chans or not (np.array(load_channels) == np.array(chans)).all():
from utime.errors import ChannelNotFoundError
raise ChannelNotFoundError("Could not load channels {} "
"from file {}. "
"Actually loaded: {}".format(
load_channels, file_path, chans))
return rec
row['signals'] = record.p_signal
row['signal_length'] = record.sig_len
channels = record.sig_name
signals = record.p_signal.transpose()
row['channels'] = channels
for channel, signal in zip(channels, signals):
row[channel] = signal
return row
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(record_name=os.path.join('data', record_name))
label = comments_to_dict(record.comments)['Reason for admission'][1:]
patient = record_name.split('/')[0]
signal_length = record.sig_len
records.append({'name':record_name, 'label':label, 'patient':patient, 'signal_length':signal_length})
channels = record.sig_name
df_records = pd.DataFrame(records)
labels = df_records['label'].unique()
df_records['label'].value_counts()
selected_labels = [
'Healthy control',
from tqdm import tqdm
from wfdb import io, plot
import pandas as pd
# https://mc.ai/diagnosing-myocardial-infarction-using-long-short-term-memory-networks-lstms/
# The folder where you want to store your data
data_folder = '../data/'
# First get the list of available records and then download
# those records and store them in data_folder.
record_names = io.get_record_list('sddb')
#io.dl_database('sddb', data_folder, record_names)
# Read the first record
record_name = record_names[0]
print(record_name)
record = io.rdrecord(record_name="../data/" + record_name)
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(record_name=os.path.join('../data/', record_name))
print(record.sig_name)
print(record.comments)
# label = comments_to_dict(record.comments)['Reason for admission'][1:]
label = record.comments
patient = record_name.split('/')[0]
signal_length = record.sig_len
records.append({
'name': record_name,
'label': label,
'patient': patient,
'signal_length': signal_length
def read_ecg(file_path):
#read ecg
record = io.rdrecord(file_path)
return record
from tqdm import tqdm
from wfdb import io, plot
import pandas as pd
# https://mc.ai/diagnosing-myocardial-infarction-using-long-short-term-memory-networks-lstms/
# The folder where you want to store your data
data_folder = '../data/'
# First get the list of available records and then download
# those records and store them in data_folder.
record_names = io.get_record_list('ptbdb')
#io.dl_database('ptbdb', data_folder, record_names)
# Read the first record
record_name = record_names[0]
print(record_name)
record = io.rdrecord(record_name="../data/ptdb/" + record_name)
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(
record_name=os.path.join('../data/ptdb/', record_name))
print(record.sig_name)
# label = comments_to_dict(record.comments)['Reason for admission'][1:]
label = record.comments[4].split(": ")[-1]
patient = record_name.split('/')[0]
signal_length = record.sig_len
records.append({
'name': record_name,
'label': label,
'patient': patient,
'signal_length': signal_length
from keras.layers import CuDNNLSTM
from sklearn.utils import shuffle
from sklearn.metrics import confusion_matrix, classification_report
from wfdb import io, plot
# The folder where you want to store your data
data_folder = '../../../data/ptdb/'
# First get the list of available records and then download
# those records and store them in data_folder.
record_names = io.get_record_list('ptbdb')
# Read the first record
record_name = record_names[0]
record = io.rdrecord(record_name=os.path.join(data_folder, record_name))
records = []
for record_name in tqdm(record_names):
record = io.rdrecord(record_name=os.path.join(data_folder, record_name))
# label = comments_to_dict(record.comments)['Reason for admission'][1:]
print(record.samps_per_frame)
label = record.comments[4].split(": ")[-1]
patient = record_name.split('/')[0]
signal_length = record.sig_len
records.append({
'name': record_name,
'label': label,
'patient': patient,
'signal_length': signal_length
})