Feature extraction update #2
Conversation
tutorials/seis_feature.py
Outdated
| env = compute_envelope(tr.data) | ||
|
|
||
| if envfilter == True: | ||
| sos = signal.butter(1, 0.01, 'lp', fs = tr.stats.sampling_rate, output = 'sos') |
There was a problem hiding this comment.
i see here that the bounds of the filter are hardcoded in the function. would you consider moving these frequencies out of the function as argument? they could belong to a dictionary so that they remain specific to each function but also general enough that users can play with these frequency bands without having the change the functions
tutorials/seis_feature.py
Outdated
| @@ -55,6 +77,8 @@ def compute_hibert(tr, env): | |||
| auto = np.correlate(tr.data, tr.data, 'same') ## autocorrelation function | |||
|
|
|||
| t = tr.times() | |||
|
|
|||
| # family number 1: Based on waveforms. | |||
| all_attr[0,0] = t[-1] - t[0] # Duration | |||
There was a problem hiding this comment.
is this duration? that seems to be just the window length
tutorials/seis_feature.py
Outdated
| sos_5_70 = signal.butter(N = 2, Wn= [3,10], btype = 'bp', fs=tr.stats.sampling_rate, output='sos') | ||
| sos_50_100 = signal.butter(N = 2, Wn= [10,20], btype = 'bp', fs=tr.stats.sampling_rate, output='sos') | ||
| sos_5_100 = signal.butter(N = 2, Wn= [20,50], btype = 'bp', fs=tr.stats.sampling_rate, output='sos') | ||
| sos_0p1_1 = signal.butter(N = 2, Wn= [0.1,1], btype = 'bp', fs=tr.stats.sampling_rate, output='sos') |
There was a problem hiding this comment.
hardcoding frequency in these filters could be prone to bugs for other users.
tutorials/seis_feature.py
Outdated
|
|
||
| ft = ft[0:len(ft)//2] | ||
| ft = abs(np.fft.fft(tr.data)) ## Discrete Fast Fourtier Transform |
There was a problem hiding this comment.
you could potentially save 50% of the calculation time by using the scipy fft. https://stackoverflow.com/questions/6363154/what-is-the-difference-between-numpy-fft-and-scipy-fftpack
tutorials/seis_feature.py
Outdated
| all_attr = np.empty((1,NATT), dtype = float) | ||
|
|
||
| attributes = ['Duration', 'RappMaxMean', 'RappMaxMedian', 'AsDec','KurtoSig', 'KurtoEnv', 'SkewSig', | ||
| attributes = ['Window_Length', 'RappMaxMean', 'RappMaxMedian', 'AsDec','KurtoSig', 'KurtoEnv', 'SkewSig', | ||
| 'SkewEnv', 'CorPeakNumber', 'Energy1/3Cor', 'Energy2/3Cor', 'int_ratio','E_0.1_1','E_1_3', | ||
| 'E_3_10', 'E_10_20','E_20_50','Kurt_0.1_1','Kurt_1_3','Kurt_3_10','Kurt_10_20','Kurt_20_50', |
There was a problem hiding this comment.
these have the hard coded frequency name, i forgot to notice that. it may feel less explanatory, but maybe using E_f1_f2, E_f2_f3, ... would work
tutorials/seis_feature.py
Outdated
| starttime = tr.stats.starttime | ||
|
|
||
|
|
||
| tr.trim(starttime, starttime + 239.95) |
There was a problem hiding this comment.
could you trim after removing the instrumental response?
No description provided.