def getMetrics(trace):
data = trace.data
mean = data.mean()
median = np.median(data)
stdv = data.std()
maximum = np.amax(data)
trace.taper(type='hamming', max_percentage=0.05, max_length=5)
data = trace.data
repFreq = central_frequency_unwindowed(data, df)
filtered = bandpass(data, 0.01, 1.5625, df)
sumEnergy = np.sum(
welch(filtered, np.hamming(len(data)), next_pow_2(len(data))))
return [mean, median, stdv, maximum, repFreq, sumEnergy]
def preprocess(df):
df.drop([
'receiver_latitude', 'receiver_longitude', 'receiver_elevation_m',
'p_arrival_sample', 'p_travel_sec', 's_arrival_sample',
'source_origin_time', 'source_latitude', 'source_longitude',
'source_depth_km'
],
axis=1,
inplace=True)
# Degree of rectiliniarity (polarization)
print(' Processing - Degree of rectiliniarity (polarization)')
df['rect_azimuth'], df['rect_incidence'], df['rect_rectilinearity'], df[
'rect_planarity'] = zip(
*df.apply(lambda x: flinn([x['Z'], x['N'], x['E']]), axis=1))
# trace-by-trace features
print(' Starting Trace-By-Trace feature processing')
trace_list = ['E', 'N', 'Z']
for tl in trace_list:
# SPECTRAL CENTROID
print(' Processing Spectral Centroid')
df['spectral_centroid_{}'.format(tl)] = df[tl].apply(
_spectral_centroid)
# RMS of frequency amplitude
print(' RMS of frequency amplitude')
df['rms_freq_amp_{}'.format(tl)] = df[tl].apply(
lambda x: np.sqrt(np.mean(np.square(np.real(np.fft.fft(x))))))
# Maximum power of frequency amplitude
print(' Maximum power of frequency amplitude')
df['max_power_freq_amp_{}'.format(tl)] = df[tl].apply(
lambda x: np.sqrt(
signal.periodogram(x, 100, 'flattop', scaling='spectrum')[1].
max()))
# Dominant frequency
print(' Dominant frequency')
df['dominant_freq_{}'.format(tl)] = df[tl].apply(
lambda x: central_frequency_unwindowed(x, fs=100))
# return trace_id and model columns
return df[[
'trace_id', 'snr_db_E', 'snr_db_N', 'snr_db_Z', 'spectral_centroid_E',
'spectral_centroid_N', 'spectral_centroid_Z', 'rect_azimuth',
'rect_incidence', 'rect_rectilinearity', 'rect_planarity',
'rms_freq_amp_E', 'rms_freq_amp_N', 'rms_freq_amp_Z',
'max_power_freq_amp_E', 'max_power_freq_amp_N', 'max_power_freq_amp_Z',
'dominant_freq_E', 'dominant_freq_N', 'dominant_freq_Z'
]]
def plotBandSpec(trace, mode='save', low=24.99, high=0.001):
#copy the data
df = 50
#demean, detrend, bandpass filter
trace.detrend('demean')
trace.detrend('linear')
trace.filter('bandpass',
freqmin=high,
freqmax=low,
corners=2,
zerophase=True)
traceCopy = trace.copy()
data = traceCopy.data
repFreq = central_frequency_unwindowed(data, df)
bw = bandwidth(data, df)
print("Central frequency: " + str(repFreq) + "\nBandwidth: " + str(bw))
def preprocess(df):
# Rolling AVG traces
print('Calculate rolling Avg')
df['E_MA250'] = df['E'].apply(lambda x: moving_average(x, 250))
df['N_MA250'] = df['N'].apply(lambda x: moving_average(x, 250))
df['Z_MA250'] = df['Z'].apply(lambda x: moving_average(x, 250))
df['E_MA1000'] = df['E'].apply(lambda x: moving_average(x, 1000))
df['N_MA1000'] = df['N'].apply(lambda x: moving_average(x, 1000))
df['Z_MA1000'] = df['Z'].apply(lambda x: moving_average(x, 1000))
# Degree of rectiliniarity (polarization)
print(' Processing - Degree of rectiliniarity (polarization)')
df['rect_azimuth'], df['rect_incidence'], df['rect_rectilinearity'], df[
'rect_planarity'] = zip(
*df.apply(lambda x: flinn([x['Z'], x['N'], x['E']]), axis=1))
df['rect_azimuth_MA250'], df['rect_incidence_MA250'], df[
'rect_rectilinearity_MA250'], df['rect_planarity_MA250'] = zip(
*df.apply(
lambda x: flinn([x['Z_MA250'], x['N_MA250'], x['E_MA250']]),
axis=1))
df['rect_azimuth_MA1000'], df['rect_incidence_MA1000'], df[
'rect_rectilinearity_MA1000'], df['rect_planarity_MA1000'] = zip(
*df.apply(
lambda x: flinn([x['Z_MA1000'], x['N_MA1000'], x['E_MA1000']]),
axis=1))
# trace-by-trace features
print(' Starting Trace-By-Trace feature processing')
trace_list = ['E', 'N', 'Z']
# Envelope similarity
print(' Processing - Envelope Similarity')
for tl in trace_list:
df['{}_env_sim_deep_max'.format(tl)], df['{}_env_sim_deep_mean'.format(
tl)], df['{}_env_sim_shallow_max'.format(tl)], df[
'{}_env_sim_shallow_mean'.format(tl)] = zip(
*df[tl].apply(lambda x: envelope_similarity(x)))
trace_list_MA = [
'E', 'N', 'Z', 'E_MA250', 'N_MA250', 'Z_MA250', 'E_MA1000', 'N_MA1000',
'Z_MA1000'
]
for tl in trace_list_MA:
# SPECTRAL CENTROID
print(' Processing Spectral Centroid - {}'.format(tl))
df['spectral_centroid_{}'.format(tl)] = df[tl].apply(
_spectral_centroid)
# RMS of frequency amplitude
print(' RMS of frequency amplitude - {}'.format(tl))
df['rms_freq_amp_{}'.format(tl)] = df[tl].apply(
lambda x: np.sqrt(np.mean(np.square(np.real(np.fft.fft(x))))))
# Maximum power of frequency amplitude
print(' Maximum power of frequency amplitude - {}'.format(tl))
df['max_power_freq_amp_{}'.format(tl)] = df[tl].apply(
lambda x: np.sqrt(
signal.periodogram(x, 100, 'flattop', scaling='spectrum')[1].
max()))
# Dominant frequency
print(' Dominant frequency - {}'.format(tl))
df['dominant_freq_{}'.format(tl)] = df[tl].apply(
lambda x: central_frequency_unwindowed(x, fs=100))
# Waveform correlation
print(' Waveform Correlation - {}'.format(tl))
df['xcor_{}_deep_max'.format(tl)], df['xcor_{}_deep_mean'.format(
tl)], df['xcor_{}_shallow_max'.format(tl)], df[
'xcor_{}_shallow_mean'.format(tl)] = zip(
*df[tl].apply(lambda x: waveform_xc_properties(x)))
# return trace_id and model columns
return df[[
'trace_id', 'spectral_centroid_E', 'spectral_centroid_N',
'spectral_centroid_Z', 'spectral_centroid_E_MA250',
'spectral_centroid_N_MA250', 'spectral_centroid_Z_MA250',
'spectral_centroid_E_MA1000', 'spectral_centroid_N_MA1000',
'spectral_centroid_Z_MA1000', 'rect_azimuth', 'rect_incidence',
'rect_rectilinearity', 'rect_planarity', 'rect_azimuth_MA250',
'rect_incidence_MA250', 'rect_rectilinearity_MA250',
'rect_planarity_MA250', 'rect_azimuth_MA1000', 'rect_incidence_MA1000',
'rect_rectilinearity_MA1000', 'rect_planarity_MA1000',
'rms_freq_amp_E', 'rms_freq_amp_N', 'rms_freq_amp_Z',
'rms_freq_amp_E_MA250', 'rms_freq_amp_N_MA250', 'rms_freq_amp_Z_MA250',
'rms_freq_amp_E_MA1000', 'rms_freq_amp_N_MA1000',
'rms_freq_amp_Z_MA1000', 'max_power_freq_amp_E',
'max_power_freq_amp_N', 'max_power_freq_amp_Z',
'max_power_freq_amp_E_MA250', 'max_power_freq_amp_N_MA250',
'max_power_freq_amp_Z_MA250', 'max_power_freq_amp_E_MA1000',
'max_power_freq_amp_N_MA1000', 'max_power_freq_amp_Z_MA1000',
'dominant_freq_E', 'dominant_freq_N', 'dominant_freq_Z',
'dominant_freq_E_MA250', 'dominant_freq_N_MA250',
'dominant_freq_Z_MA250', 'dominant_freq_E_MA1000',
'dominant_freq_N_MA1000', 'dominant_freq_Z_MA1000', 'xcor_E_deep_max',
'xcor_E_deep_mean', 'xcor_E_shallow_max', 'xcor_E_shallow_mean',
'xcor_N_deep_max', 'xcor_N_deep_mean', 'xcor_N_shallow_max',
'xcor_N_shallow_mean', 'xcor_Z_deep_max', 'xcor_Z_deep_mean',
'xcor_Z_shallow_max', 'xcor_Z_shallow_mean', 'xcor_E_MA250_deep_max',
'xcor_E_MA250_deep_mean', 'xcor_E_MA250_shallow_max',
'xcor_E_MA250_shallow_mean', 'xcor_N_MA250_deep_max',
'xcor_N_MA250_deep_mean', 'xcor_N_MA250_shallow_max',
'xcor_N_MA250_shallow_mean', 'xcor_Z_MA250_deep_max',
'xcor_Z_MA250_deep_mean', 'xcor_Z_MA250_shallow_max',
'xcor_Z_MA250_shallow_mean', 'xcor_E_MA1000_deep_max',
'xcor_E_MA1000_deep_mean', 'xcor_E_MA1000_shallow_max',
'xcor_E_MA1000_shallow_mean', 'xcor_N_MA1000_deep_max',
'xcor_N_MA1000_deep_mean', 'xcor_N_MA1000_shallow_max',
'xcor_N_MA1000_shallow_mean', 'xcor_Z_MA1000_deep_max',
'xcor_Z_MA1000_deep_mean', 'xcor_Z_MA1000_shallow_max',
'xcor_Z_MA1000_shallow_mean', 'E_env_sim_deep_max',
'E_env_sim_deep_mean', 'E_env_sim_shallow_max',
'E_env_sim_shallow_mean', 'N_env_sim_deep_max', 'N_env_sim_deep_mean',
'N_env_sim_shallow_max', 'N_env_sim_shallow_mean',
'Z_env_sim_deep_max', 'Z_env_sim_deep_mean', 'Z_env_sim_shallow_max',
'Z_env_sim_shallow_mean'
]]