else:
f = data
#f = f - np.mean(f)
N = 3 #number of supports
detect = "locmax" #detection mode: locmax, locmaxmin, locmaxminf
reg = 'none' #spectrum regularization - it is smoothed with an average (or gaussian) filter
lengthFilter = 0 #length or average or gaussian filter
sigmaFilter = 0 #sigma of gaussian filter
Fs = 1 #sampling frequency, in Hz (if unknown, set 1)
ewt, mfb ,boundaries = ewtpy.EWT1D(f,
N = N,
log = 0,
detect = detect,
completion = 0,
reg = reg,
lengthFilter = lengthFilter,
sigmaFilter = sigmaFilter)
#plot original signal and decomposed modes
plt.figure()
plt.subplot(211)
plt.plot(f)
plt.title('Original signal %s'%signal)
plt.subplot(212)
plt.plot(ewt)
plt.title('EWT modes')
#%% show boundaries
ff = np.fft.fft(f)
import numpy as np import matplotlib.pyplot as plt import ewtpy T = 1000 t = np.arange(1,T+1)/T f = np.cos(2*np.pi*0.8*t) + 2*np.cos(2*np.pi*10*t)+0.8*np.cos(2*np.pi*100*t) ewt, mfb ,boundaries = ewtpy.EWT1D(f, N = 3) plt.plot(f) plt.plot(ewt) plt.show()
#%% Fig 4: EWT spectrum segmentation
Fs = 173.61 #EEG sampling frequency
#EWT parameters
FFTreg = 'gaussian'
FFTregLen = 25
gaussSigma = 5
plt.rcParams.update({'font.size': 8})
f1 = np.loadtxt("BonnDataset/data/S/S013.txt")
b, a = signal.butter(4, 40 / (0.5 * Fs), btype='low', analog=False)
f1 = signal.filtfilt(b, a, f1)
f1 = f1 - np.mean(f1)
ewt, mfb, boundaries = ewtpy.EWT1D(f1,
N=5,
log=0,
detect="locmax",
completion=0,
reg=FFTreg,
lengthFilter=FFTregLen,
sigmaFilter=gaussSigma)
ff = np.fft.fft(f1)
ff = abs(ff[:ff.size // 2]) #one-sided magnitude
regFilter = np.zeros(FFTregLen)
regFilter[regFilter.size //
2] = 1 #prefer odd filter lengths - otherwise the gaussian is skewed
presig = np.convolve(ff, gaussian_filter(regFilter, gaussSigma), mode='same')
fftFreq = np.fft.fftfreq(f1.size, 1 / Fs)
fftFreq = fftFreq[:fftFreq.size // 2]
plt.figure(figsize=(3.54, 2.5))
ax = plt.subplot(111)
#ax.plot(fftFreq,ff)
ax.plot(fftFreq, presig, 'k')
def fun_loadExFeats(dSet,
ri,
idx,
item,
Fs,
LPcutoff,
Nmodes,
FFTregLen=25,
gaussSigma=5,
FFTreg='gaussian'):
#load eeg
featNames = ["Group", "decTime"]
featsTuple = {
"EMD": 0,
"EEMD": 0,
"CEEMDAN": 0,
"EWT": 0,
"VMD": 0,
"Orig": 0
}
if dSet == "NSC_ND":
fLoad = loadmat("%s/data/%s/%s%d" % (dSet, item, item, ri + 1))
f = fLoad[item][:, 0]
ltemp = int(np.ceil(f.size / 2)) #to behave the same as matlab's round
fMirr = np.append(np.flip(f[0:ltemp - 1], axis=0), f)
fMirr = np.append(fMirr, np.flip(f[-ltemp - 1:-1], axis=0))
f = np.copy(fMirr)
if dSet == "BonnDataset":
f = np.loadtxt("%s/data/%s/%s%.3d.txt" % (dSet, item, item, ri + 1))
#preprocessing - LP filter and remove DC
f = f - np.mean(f)
b, a = signal.butter(4, LPcutoff / (0.5 * Fs), btype='low', analog=False)
fp = signal.filtfilt(b, a, f)
#% EMD features
tic = time.time()
emd = EMD()
emd.MAX_ITERATION = 2000
IMFs = emd.emd(fp, max_imf=Nmodes)
toc = time.time()
featsTuple["EMD"] = toc - tic #execution time (decomposition)
if Nmodes != IMFs.shape[0] - 1:
print("\nCheck number of EMD modes: %s%.3d" % (item, ri + 1))
for mi in range(IMFs.shape[0]):
featOut, labelTemp = featExtract(IMFs[mi, :], Fs, welchWin=1024)
featsTuple["EMD"] = np.append(featsTuple["EMD"], featOut)
#write feature name header
if ri == 0 and idx == 0:
for ii in labelTemp:
featNames = np.append(featNames, "%s%d" % (ii, mi))
if IMFs.shape[0] < Nmodes + 1:
featsTuple["EMD"] = np.append(
featsTuple["EMD"], np.zeros(Nfeats * (Nmodes + 1 - IMFs.shape[0])))
#% EEMD - Ensemble Empirical Mode Decomposition
tic = time.time()
if __name__ == "__main__":
eemd = EEMD(trials=200)
eemd.MAX_ITERATION = 2000
eIMFs = eemd(fp, max_imf=Nmodes)
toc = time.time()
featsTuple["EEMD"] = toc - tic #execution time (decomposition )
if Nmodes != eIMFs.shape[0] - 1:
print("\nCheck number of EEMD modes: %s%.3d" % (item, ri + 1))
#for each mode, extract features
for mi in range(eIMFs.shape[0]):
featOut, labelTemp = featExtract(eIMFs[mi, :], Fs, welchWin=1024)
featsTuple["EEMD"] = np.append(featsTuple["EEMD"], featOut)
if eIMFs.shape[0] < Nmodes + 1:
featsTuple["EEMD"] = np.append(
featsTuple["EEMD"],
np.zeros(Nfeats * (Nmodes + 1 - eIMFs.shape[0])))
#% CEEMDAN - Complete Ensemble Empirical Mode Decomposition with Adaptive Noise
tic = time.time()
if __name__ == "__main__":
ceemdan = CEEMDAN()
ceIMFs = ceemdan(fp, max_imf=Nmodes)
toc = time.time()
featsTuple["CEEMDAN"] = toc - tic #execution time (decomposition )
if Nmodes != ceIMFs.shape[0] - 1:
print("\nCheck number of CEEMDAN modes: %s%.3d" % (item, ri + 1))
#for each mode, extract features
for mi in range(ceIMFs.shape[0]):
featOut, labelTemp = featExtract(ceIMFs[mi, :], Fs, welchWin=1024)
featsTuple["CEEMDAN"] = np.append(featsTuple["CEEMDAN"], featOut)
if ceIMFs.shape[0] < Nmodes + 1:
featsTuple["CEEMDAN"] = np.append(
featsTuple["CEEMDAN"],
np.zeros(Nfeats * (Nmodes + 1 - ceIMFs.shape[0])))
#%EWT features
tic = time.time()
ewt, _, _ = ewtpy.EWT1D(fp,
N=Nmodes,
log=0,
detect="locmax",
completion=0,
reg=FFTreg,
lengthFilter=FFTregLen,
sigmaFilter=gaussSigma)
toc = time.time()
featsTuple["EWT"] = toc - tic #execution time (decomposition )
if Nmodes != ewt.shape[1]:
print("\nCheck number of EWT modes: %s%.3d" % (item, ri + 1))
#for each mode, extract features
for mi in range(Nmodes):
featOut, labelTemp = featExtract(ewt[:, mi], Fs, welchWin=1024)
featsTuple["EWT"] = np.append(featsTuple["EWT"], featOut)
#% VMD features
DC = np.mean(fp) # no DC part imposed
tic = time.time()
vmd, _, _ = VMD(fp, alpha, tau, Nmodes, DC, init, tol)
toc = time.time()
featsTuple["VMD"] = toc - tic #execution time (decomposition )
if Nmodes != vmd.shape[0]:
print("\nCheck number of VMD modes: %s%.3d" % (item, ri + 1))
#for each mode, extract features
for mi in range(Nmodes):
featOut, labelTemp = featExtract(vmd[mi, :], Fs, welchWin=1024)
featsTuple["VMD"] = np.append(featsTuple["VMD"], featOut)
#% Original non-decomposed signal features
tic = time.time()
featOut, labelTemp = featExtract(fp, Fs, welchWin=1024)
toc = time.time()
featsTuple["Orig"] = np.append(toc - tic, featOut)
return item, featsTuple