#!/usr/bin/env python
# -*- coding: utf-8 -*-
'''
Please see this before you run this file!!!!!!
NOTICE:
Be advised that this “Extract-Raw-Data-Into-Matlab-Files.py” Python File
should only be executed under the Python 2 Environment.
I highly recommend to execute the file under the Python 2.7 Environment
because I have passed the test.
However, if you are using Python 3 Environment to run this file,
I'm afraid there will be an error and the generated labels will be wrong.
If you have any question, please be sure to let me know.
My email is shuyuej@ieee.org.
Thanks a lot.
'''
import numpy as np
import io
import os
import pyedflib
import scipy.io as sio
from scipy.signal import butter, filtfilt, iirnotch, lfilter
MOVEMENT_START = 1 * 160 # MI starts 1s after trial begin
MOVEMENT_END = 5 * 160 # MI lasts 4 seconds
NOISE_LEVEL = 0.01
PHYSIONET_ELECTRODES = {
1: "FC5", 2: "FC3", 3: "FC1", 4: "FCz", 5: "FC2", 6: "FC4",
7: "FC6", 8: "C5", 9: "C3", 10: "C1", 11: "Cz", 12: "C2",
13: "C4", 14: "C6", 15: "CP5", 16: "CP3", 17: "CP1", 18: "CPz",
19: "CP2", 20: "CP4", 21: "CP6", 22: "Fp1", 23: "Fpz", 24: "Fp2",
25: "AF7", 26: "AF3", 27: "AFz", 28: "AF4", 29: "AF8", 30: "F7",
31: "F5", 32: "F3", 33: "F1", 34: "Fz", 35: "F2", 36: "F4",
37: "F6", 38: "F8", 39: "FT7", 40: "FT8", 41: "T7", 42: "T8",
43: "T9", 44: "T10", 45: "TP7", 46: "TP8", 47: "P7", 48: "P5",
49: "P3", 50: "P1", 51: "Pz", 52: "P2", 53: "P4", 54: "P6",
55: "P8", 56: "PO7", 57: "PO3", 58: "POz", 59: "PO4", 60: "PO8",
61: "O1", 62: "Oz", 63: "O2", 64: "Iz"}
def load_edf_signals(path):
try:
sig = pyedflib.EdfReader(path)
n = sig.signals_in_file
signal_labels = sig.getSignalLabels()
sigbuf = np.zeros((n, sig.getNSamples()[0]))
for j in np.arange(n):
sigbuf[j, :] = sig.readSignal(j)
# (n,3) annotations: [t in s, duration, type T0/T1/T2]
annotations = sig.read_annotation()
except KeyboardInterrupt:
# prevent memory leak and access problems of unclosed buffers
sig._close()
raise
sig._close()
del sig
return sigbuf.transpose(), annotations
def get_physionet_electrode_positions():
refpos = get_electrode_positions()
return np.array([refpos[PHYSIONET_ELECTRODES[idx]] for idx in range(1, 65)])
def projection_2d(loc):
"""
Azimuthal equidistant projection (AEP) of 3D carthesian coordinates.
Preserves distance to origin while projecting to 2D carthesian space.
loc: N x 3 array of 3D points
returns: N x 2 array of projected 2D points
"""
x, y, z = loc[:, 0], loc[:, 1], loc[:, 2]
theta = np.arctan2(y, x) # theta = azimuth
rho = np.pi / 2 - np.arctan2(z, np.hypot(x, y)) # rho = pi/2 - elevation
return np.stack((np.multiply(rho, np.cos(theta)), np.multiply(rho, np.sin(theta))), 1)
def get_electrode_positions():
"""
Returns a dictionary (Name) -> (x,y,z) of electrode name in the extended
10-20 system and its carthesian coordinates in unit sphere.
"""
positions = dict()
with io.open("electrode_positions.txt", "r") as pos_file:
for line in pos_file:
parts = line.split()
positions[parts[0]] = tuple([float(part) for part in parts[1:]])
return positions
def load_physionet_data(subject_id, num_classes=4, long_edge=False):
"""
subject_id: ID (1-109) for the subject to be loaded from file
num_classes: number of classes (2, 3 or 4) for L/R, L/R/0, L/R/0/F
long_edge: if False include 1s before and after MI, if True include 3s
returns (X, y, pos, fs)
X: Trials with shape (N_subjects, N_trials, N_samples, N_channels)
y: labels with shape (N_subjects, N_trials, N_classes)
pos: 2D projected electrode positions
fs: sample rate
"""
SAMPLE_RATE = 160
EEG_CHANNELS = 64
BASELINE_RUN = 1
MI_RUNS = [4, 8, 12] # l/r fist
if num_classes >= 4:
MI_RUNS += [6, 10, 14] # feet (& fists)
# total number of samples per long run
RUN_LENGTH = 125 * SAMPLE_RATE
# length of single trial in seconds
TRIAL_LENGTH = 4 if not long_edge else 6
NUM_TRIALS = 21 * num_classes
n_runs = len(MI_RUNS)
X = np.zeros((n_runs, RUN_LENGTH, EEG_CHANNELS))
events = []
base_path = 'S%03dR%02d.edf'
for i_run, current_run in enumerate(MI_RUNS):
# load from file
path = base_path % (subject_id, current_run)
signals, annotations = load_edf_signals(path)
X[i_run, :signals.shape[0], :] = signals
# read annotations
current_event = [i_run, 0, 0, 0] # run, class (l/r), start, end
for annotation in annotations:
t = int(annotation[0] * SAMPLE_RATE * 1e-7)
action = int(annotation[2][1])
if action == 0 and current_event[1] != 0:
# make 6 second runs by extending snippet
length = TRIAL_LENGTH * SAMPLE_RATE
pad = (length - (t - current_event[2])) / 2
current_event[2] -= pad + (t - current_event[2]) % 2
current_event[3] = t + pad
if (current_run - 6) % 4 != 0 or current_event[1] == 2:
if (current_run - 6) % 4 == 0:
current_event[1] = 3
events.append(current_event)
elif action > 0:
current_event = [i_run, action, t, 0]
# split runs into trials
num_mi_trials = len(events)
trials = np.zeros((NUM_TRIALS, TRIAL_LENGTH * SAMPLE_RATE, EEG_CHANNELS))
labels = np.zeros((NUM_TRIALS, num_classes))
for i, ev in enumerate(events):
trials[i, :, :] = X[ev[0], ev[2]:ev[3]]
labels[i, ev[1] - 1] = 1.
if num_classes < 3:
return (trials[:num_mi_trials, ...],
labels[:num_mi_trials, ...],
projection_2d(get_physionet_electrode_positions()),
SAMPLE_RATE)
else:
# baseline run
path = base_path % (subject_id, BASELINE_RUN)
signals, annotations = load_edf_signals(path)
SAMPLES = TRIAL_LENGTH * SAMPLE_RATE
for i in range(num_mi_trials, NUM_TRIALS):
offset = np.random.randint(0, signals.shape[0] - SAMPLES)
trials[i, :, :] = signals[offset: offset+SAMPLES, :]
labels[i, -1] = 1.
return trials, labels, projection_2d(get_physionet_electrode_positions()), SAMPLE_RATE
def load_raw_data(electrodes, subject=None, num_classes=4, long_edge=False):
# load from file
trials = []
labels = []
if subject == None:
subject_ids = range(11, 15)
else:
try:
subject_ids = [int(subject)]
except:
subject_ids = subject
for subject_id in subject_ids:
try:
t, l, loc, fs = load_physionet_data(subject_id, num_classes, long_edge=long_edge)
if num_classes == 2 and t.shape[0] != 42:
# drop subjects with less trials
continue
trials.append(t[:, :, electrodes])
labels.append(l)
except:
pass
return np.array(trials, dtype=np.float64).reshape((len(trials),) + trials[0].shape + (1,)), \
np.array(labels, dtype=np.float64)
def butter_bandpass(lowcut, highcut, fs, order=5):
nyq = 0.5 * fs
low = lowcut / nyq
high = highcut / nyq
b, a = butter(order, [low, high], btype='band')
return b, a
def butter_bandpass_filter(data, lowcut, highcut, fs, order=5):
b, a = butter_bandpass(lowcut, highcut, fs, order=order)
y = filtfilt(b, a, data)
return y
print("Start to save the Files!")
# Sample rate and desired cutoff frequencies (in Hz).
fs = 160.0
lowcut = 5.0
highcut = 30.0
# Save Dataset of 4 clases
nclasses = 4
# Save 20 subjects' dataset
SAVE = '20-Subjects'
if not os.path.exists(SAVE):
os.mkdir(SAVE)
subject = range(1, 21)
# Save 64 electrodes
for i in range(0, 64):
electrodes = [i]
X = 'X_' + str(i)
Y = 'Y_' + str(i)
X, Y = load_raw_data(electrodes=electrodes, subject=subject, num_classes=nclasses)
X = np.squeeze(X)
sio.savemat(SAVE + 'Dataset_%d.mat' % int(i+1), {'Dataset': X})
sio.savemat(SAVE + 'Labels_%d.mat' % int(i+1), {'Labels': Y})
print("Finished saving %d electrodes" % int(i+1))
# SAVE = '100-Subjects/'
#
# if not os.path.exists(SAVE):
# os.mkdir(SAVE)
#
# subject = range(1, 102)
#
# for i in range(0, 64):
# electrodes = [i]
# X = 'X_' + str(i)
# Y = 'Y_' + str(i)
# X, Y = load_raw_data(electrodes=electrodes, subject=subject, num_classes=nclasses)
# X = np.squeeze(X)
#
# # # Notch Filter
# # b, a = iirnotch(w0=60.0, Q=30.0, fs=fs)
# # X = lfilter(b, a, X)
# #
# # # Butterworth Band-pass filter
# # X = butter_bandpass_filter(X, lowcut, highcut, fs, order=4)
#
# sio.savemat(SAVE + 'Dataset_%d.mat' % int(i+1), {'Dataset': X})
# sio.savemat(SAVE + 'Labels_%d.mat' % int(i+1), {'Labels': Y})
#
# print("Finished saving %d electrodes" % int(i+1))
Datasets
Models
