def extract_hash_features(list_features,date,file,dic,permut_file,plot=False):
"""
Extracts hash table values.
"""
from fingerprint_functions import FuncFingerprint, spectrogram, ponset_grad, vec_compute_signature, LSH
s = SeismicTraces(file,utcdatetime.UTCDateTime(str(date)))
list_attr = s.__dict__.keys()
if 'tr_grad' not in list_attr:
for feat in list_features:
dic[feat] = np.nan
return dic
full_tr = s.tr
grad = s.tr_grad
q = [100.,.8,1]
(full_spectro,f,full_time,end) = spectrogram(full_tr,param=q)
ponset = ponset_grad(full_tr,grad,full_time,plot=plot)
idebut = ponset-int(2*full_spectro.shape[1]/full_time[-1])
print ponset, idebut
if idebut < 0:
idebut = 0
ifin = idebut+full_spectro.shape[0]
time = full_time[idebut:ifin]
spectro = full_spectro[:,idebut:ifin]
if spectro.shape[1] < spectro.shape[0]:
spectro = full_spectro[:,-spectro.shape[0]:]
haar_bin = FuncFingerprint(np.log10(spectro),time,full_tr,f,end,plot=plot,error=plot)
m = haar_bin.shape[1] # number of columns
n = haar_bin.shape[0] # number of rows
vec_bin = np.zeros(m*n)
for i in range(n):
for j in range(m):
vec_bin[m*i+j] = haar_bin[i,j]
# HASH TABLES
pp = 500
p_file = '%s_%d'%(permut_file,full_spectro.shape[0])
if not os.path.isfile(p_file):
print "Permutation"
from fingerprint_functions import define_permutation
import cPickle
permut = define_permutation(len(vec_bin),pp)
with open(p_file,'wb') as file:
my_pickler = cPickle.Pickler(file)
my_pickler.dump(permut)
file.close()
else:
import cPickle
with open(p_file,'rb') as file:
my_depickler = cPickle.Unpickler(file)
permut = my_depickler.load()
file.close()
MH_sign = vec_compute_signature(vec_bin,permut)
HashTab = LSH(MH_sign,l=50)
for iht,ht in enumerate(HashTab):
dic['%d'%iht] = ht
if plot:
plt.show()
return dic
def extract_hash_features(list_features,
date,
file,
dic,
permut_file,
plot=False):
"""
Extracts hash table values.
"""
from fingerprint_functions import FuncFingerprint, spectrogram, ponset_grad, vec_compute_signature, LSH
s = SeismicTraces(file, utcdatetime.UTCDateTime(str(date)))
list_attr = s.__dict__.keys()
if 'tr_grad' not in list_attr:
for feat in list_features:
dic[feat] = np.nan
return dic
full_tr = s.tr
grad = s.tr_grad
q = [100., .8, 1]
(full_spectro, f, full_time, end) = spectrogram(full_tr, param=q)
ponset = ponset_grad(full_tr, grad, full_time, plot=plot)
idebut = ponset - int(2 * full_spectro.shape[1] / full_time[-1])
print ponset, idebut
if idebut < 0:
idebut = 0
ifin = idebut + full_spectro.shape[0]
time = full_time[idebut:ifin]
spectro = full_spectro[:, idebut:ifin]
if spectro.shape[1] < spectro.shape[0]:
spectro = full_spectro[:, -spectro.shape[0]:]
haar_bin = FuncFingerprint(np.log10(spectro),
time,
full_tr,
f,
end,
plot=plot,
error=plot)
m = haar_bin.shape[1] # number of columns
n = haar_bin.shape[0] # number of rows
vec_bin = np.zeros(m * n)
for i in range(n):
for j in range(m):
vec_bin[m * i + j] = haar_bin[i, j]
# HASH TABLES
pp = 500
p_file = '%s_%d' % (permut_file, full_spectro.shape[0])
if not os.path.isfile(p_file):
print "Permutation"
from fingerprint_functions import define_permutation
import cPickle
permut = define_permutation(len(vec_bin), pp)
with open(p_file, 'wb') as file:
my_pickler = cPickle.Pickler(file)
my_pickler.dump(permut)
file.close()
else:
import cPickle
with open(p_file, 'rb') as file:
my_depickler = cPickle.Unpickler(file)
permut = my_depickler.load()
file.close()
MH_sign = vec_compute_signature(vec_bin, permut)
HashTab = LSH(MH_sign, l=50)
for iht, ht in enumerate(HashTab):
dic['%d' % iht] = ht
if plot:
plt.show()
return dic
def extract_norm_features(list_features,date,file,dic):
"""
Extraction of all features given by list_features, except hash
table values.
"""
s = SeismicTraces(file,utcdatetime.UTCDateTime(str(date)))
list_attr = s.__dict__.keys()
if 'tr_grad' not in list_attr:
for feat in list_features:
dic[feat] = np.nan
return dic
# Determine P-onset
s.duration()
#s.display_traces()
#s.amplitude_distribution()
if len(list_attr) > 7:
if 'Dur' in list_features:
# Mean of the predominant frequency
from waveform_features import spectrogram
s, dic['MeanPredF'], dic['TimeMaxSpec'], dic['NbPeaks'], dic['Width'], hob, vals, dic['sPredF'] = spectrogram(s,plot=False)
for i in range(len(vals)):
dic['v%d'%i] = vals[i]
dic['Dur'] = s.dur
if 'Acorr' in list_features:
from waveform_features import autocorrelation,filt_ratio
vals = autocorrelation(s,plot=False)
for i in range(len(vals)):
dic['acorr%d'%i] = vals[i]
vals = filt_ratio(s,plot=False)
for i in range(len(vals)):
dic['fratio%d'%i] = vals[i]
if 'Ene20-30' in list_features:
# Energy between 10 and 30 Hz
from waveform_features import energy_between_10Hz_and_30Hz
f1, f2 = 20,30
dic['Ene%d-%d'%(f1,f2)] = energy_between_10Hz_and_30Hz(s.tr[s.i1:s.i2]/np.max(s.tr[s.i1:s.i2]),s.dt,wd=f1,wf=f2,ponset=s.ponset-s.i1,tend=s.tend-s.i1)
if 'Ene5-10' in list_features:
f1, f2 = 5,10
dic['Ene%d-%d'%(f1,f2)] = energy_between_10Hz_and_30Hz(s.tr[s.i1:s.i2]/np.max(s.tr[s.i1:s.i2]),s.dt,wd=f1,wf=f2,ponset=s.ponset-s.i1,tend=s.tend-s.i1)
if 'Ene0-5' in list_features:
f1, f2 = .5,5
dic['Ene%d-%d'%(f1,f2)] = energy_between_10Hz_and_30Hz(s.tr[s.i1:s.i2]/np.max(s.tr[s.i1:s.i2]),s.dt,wd=f1,wf=f2,ponset=s.ponset-s.i1,tend=s.tend-s.i1)
if 'RappMaxMean' in list_features:
# Max over mean ratio of the envelope
from waveform_features import max_over_mean
dic['RappMaxMean'] = max_over_mean(s.tr[s.i1:s.i2])
if 'AsDec' in list_features:
# Ascendant phase duration over descendant phase duration
from waveform_features import growth_over_decay
p = growth_over_decay(s)
if p > 0:
dic['AsDec'] = p
if 'Growth' in list_features:
from waveform_features import desc_and_asc
dic['Growth'] = desc_and_asc(s)
if 'Skewness' in list_features:
# Skewness
from waveform_features import skewness
sk = skewness(s.tr_env[s.i1:s.i2])
dic['Skewness'] = sk
#s.amplitude_distribution()
if 'Kurto' in list_features:
# Kurtosis
from waveform_features import kurtosis_envelope
k = kurtosis_envelope(s.tr_env[s.i1:s.i2])
dic['Kurto'] = k
if ('F_low' in list_features) or ('F_up' in list_features):
# Lowest and highest frequency for kurtogram analysis
from waveform_features import kurto_bandpass
dic['F_low'],dic['F_up'] = kurto_bandpass(s,plot=False)
if 'Centroid_time' in list_features:
# Centroid time
from waveform_features import centroid_time
dic['Centroid_time'] = centroid_time(s.tr[s.i1:s.i2],s.dt,s.TF,s.ponset-s.i1,tend=s.tend-s.i1,plot=False)
if 'RappMaxMeanTF' in list_features:
# Max over mean ratio of the amplitude spectrum
from waveform_features import max_over_mean
dic['RappMaxMeanTF'] = max_over_mean(np.abs(s.TF[:len(s.TF)/2]))
if 'IFslope' in list_features:
# Average of the instantaneous frequency and slope of the unwrapped instantaneous phase
from waveform_features import instant_freq
#p,pf = instant_freq(s.tr[s.i1:s.i2],s.dt,s.TF,plot=False)
#dic['IFslope'] = np.mean((p,pf[len(pf)-1]))
vals, dic['IFslope'] = instant_freq(s.tr[s.i1:s.i2],s.dt,s.TF,s.ponset-s.i1,s.tend-s.i1,plot=False)
for i in range(len(vals)):
dic['if%d'%i] = vals[i]
if ('ibw0' in list_features) or ('Norm_envelope' in list_features):
# Average of the instantaneous frequency and normalized envelope
from waveform_features import instant_bw
vals, Nenv = instant_bw(s.tr[s.i1:s.i2],s.tr_env[s.i1:s.i2],s.dt,s.TF,s.ponset-s.i1,s.tend-s.i1,plot=False)
dic['Norm_envelope'] = Nenv
for i in range(len(vals)):
dic['ibw%d'%i] = vals[i]
if ('PredF' in list_features) or ('CentralF' in list_features) or ('Bandwidth' in list_features):
# Spectral attributes
from waveform_features import around_freq
dic['PredF'], dic['Bandwidth'], dic['CentralF'] = around_freq(s.TF,s.freqs,plot=False)
if 'Cepstrum' in list_features:
# Cepstrum
from waveform_features import cepstrum
#dic['Cepstrum'] = cepstrum(s.TF,s.freqs,plot=True)
cep = cepstrum(s.TF,s.freqs,plot=False)
dic['Ponset'] = s.ponset
return dic
def extract_norm_features(list_features, date, file, dic):
"""
Extraction of all features given by list_features, except hash
table values.
"""
s = SeismicTraces(file, utcdatetime.UTCDateTime(str(date)))
list_attr = s.__dict__.keys()
if 'tr_grad' not in list_attr:
for feat in list_features:
dic[feat] = np.nan
return dic
# Determine P-onset
s.duration()
#s.display_traces()
#s.amplitude_distribution()
if len(list_attr) > 5:
from waveform_features import spectrogram
s, dic['MeanPredF'], dic['TimeMaxSpec'], dic['NbPeaks'], dic[
'Width'], hob, vals, dic['sPredF'] = spectrogram(s, plot=False)
for i in range(len(vals)):
dic['v%d' % i] = vals[i]
dic['Dur'] = s.dur
# Compute the spectrum
s.spectrum(plot=False)
if 'Acorr' in list_features:
from waveform_features import autocorrelation, filt_ratio
vals = autocorrelation(s, plot=False)
for i in range(len(vals)):
dic['acorr%d' % i] = vals[i]
vals = filt_ratio(s, plot=False)
for i in range(len(vals)):
dic['fratio%d' % i] = vals[i]
if 'Ene20-30' in list_features:
# Energy between 10 and 30 Hz
from waveform_features import energy_between_10Hz_and_30Hz
f1, f2 = 20, 30
dic['Ene%d-%d' % (f1, f2)] = energy_between_10Hz_and_30Hz(
s.tr[s.i1:s.i2] / np.max(s.tr[s.i1:s.i2]),
s.dt,
wd=f1,
wf=f2,
ponset=s.ponset - s.i1,
tend=s.tend - s.i1)
if 'Ene5-10' in list_features:
f1, f2 = 5, 10
dic['Ene%d-%d' % (f1, f2)] = energy_between_10Hz_and_30Hz(
s.tr[s.i1:s.i2] / np.max(s.tr[s.i1:s.i2]),
s.dt,
wd=f1,
wf=f2,
ponset=s.ponset - s.i1,
tend=s.tend - s.i1)
if 'Ene0-5' in list_features:
f1, f2 = .5, 5
dic['Ene%d-%d' % (f1, f2)] = energy_between_10Hz_and_30Hz(
s.tr[s.i1:s.i2] / np.max(s.tr[s.i1:s.i2]),
s.dt,
wd=f1,
wf=f2,
ponset=s.ponset - s.i1,
tend=s.tend - s.i1)
if 'RappMaxMean' in list_features:
# Max over mean ratio of the envelope
from waveform_features import max_over_mean
dic['RappMaxMean'] = max_over_mean(s.tr[s.ponset:s.tend])
if 'AsDec' in list_features:
# Ascendant phase duration over descendant phase duration
from waveform_features import growth_over_decay
p = growth_over_decay(s)
if p > 0:
dic['AsDec'] = p
if 'Growth' in list_features:
from waveform_features import desc_and_asc
dic['Growth'] = desc_and_asc(s)
if 'Skewness' in list_features:
# Skewness
from waveform_features import skewness
sk = skewness(s.tr_env[s.ponset:s.tend])
dic['Skewness'] = sk
#s.amplitude_distribution()
if 'Kurto' in list_features:
# Kurtosis
from waveform_features import kurtosis_envelope
k = kurtosis_envelope(s.tr_env[s.ponset:s.tend])
dic['Kurto'] = k
if ('F_low' in list_features) or ('F_up' in list_features):
# Lowest and highest frequency for kurtogram analysis
from waveform_features import kurto_bandpass
dic['F_low'], dic['F_up'] = kurto_bandpass(s, plot=False)
if 'Centroid_time' in list_features:
# Centroid time
from waveform_features import centroid_time
dic['Centroid_time'] = centroid_time(s.tr[s.i1:s.i2],
s.dt,
s.TF,
s.ponset - s.i1,
tend=s.tend - s.i1,
plot=False)
if 'RappMaxMeanTF' in list_features:
# Max over mean ratio of the amplitude spectrum
from waveform_features import max_over_mean
dic['RappMaxMeanTF'] = max_over_mean(np.abs(s.TF[:len(s.TF) / 2]))
if 'IFslope' in list_features:
# Average of the instantaneous frequency and slope of the unwrapped instantaneous phase
from waveform_features import instant_freq
#p,pf = instant_freq(s.tr[s.i1:s.i2],s.dt,s.TF,plot=False)
#dic['IFslope'] = np.mean((p,pf[len(pf)-1]))
vals, dic['IFslope'] = instant_freq(s.tr[s.i1:s.i2],
s.dt,
s.TF,
s.ponset - s.i1,
s.tend - s.i1,
plot=False)
for i in range(len(vals)):
dic['if%d' % i] = vals[i]
if ('ibw0' in list_features) or ('Norm_envelope' in list_features):
# Average of the instantaneous frequency and normalized envelope
from waveform_features import instant_bw
vals, Nenv = instant_bw(s.tr[s.i1:s.i2],
s.tr_env[s.i1:s.i2],
s.dt,
s.TF,
s.ponset - s.i1,
s.tend - s.i1,
plot=False)
dic['Norm_envelope'] = Nenv
for i in range(len(vals)):
dic['ibw%d' % i] = vals[i]
if ('PredF' in list_features) or ('CentralF' in list_features) or (
'Bandwidth' in list_features):
# Spectral attributes
from waveform_features import around_freq
dic['PredF'], dic['Bandwidth'], dic['CentralF'] = around_freq(
s.TF, s.freqs, plot=False)
if 'Cepstrum' in list_features:
# Cepstrum
from waveform_features import cepstrum
#dic['Cepstrum'] = cepstrum(s.TF,s.freqs,plot=True)
cep = cepstrum(s.TF, s.freqs, plot=False)
dic['Ponset'] = s.ponset
return dic