def crosscorr_phase_angle(sig1, sig2, x, max_length=10000):
"""Return the cross correlation phase angle between 2 signals
Parameters
----------
sig1 : array
signal of length L
sig2 : array
another signal of length L
x : array
time axis for the signals sig1 and sig2
max_length : int, optional
Maximum length for the signals, signals are resampled otherwise.
Default is 10 000.
"""
assert len(sig1) == len(sig2) == len(x), \
"The signals don't have the same length."
sig_length = len(sig1)
# Resample if signal is too big thus slowing down correlation computation
if sig_length > max_length:
sig1, x = resample(sig1, max_length, x)
sig2 = resample(sig2, max_length)
sig_length = max_length
corr = np.correlate(sig1, sig2, mode="same")
xmean = sig_length/2
return float(argmax(corr) - xmean)/sig_length*x[-1] # *x[-1] to scale
def extract_data(i):
global id_legend, _id, t, intra_trace, extra_trace, img_fname
if i < n_datum and i>=0:
_id = datum.keys()[i] # 'a0' or 's1' etc.
img_fname = img_path+datum[_id][0]
igor_trace = datum[_id][1]
extra_id_0 = _id + '1_Ax1_Vm' # Axon
extra_id_1 = _id + '1_Ch2_Vm' # Dendrite
intra_id = _id + '1_Ch1_Vm'
t = igor_trace[intra_id].axis[0] * 1e6
t = t[::-1]
intra_trace = igor_trace[intra_id].data * 1e3 # unit(mV)
extra_trace = np.zeros((len(t),2))
extra_trace[:,0] = igor_trace[extra_id_0].data * 1e6 # unit(uV) Axon (left)
extra_trace[:,1] = igor_trace[extra_id_1].data * 1e6 # unit(uV) Dendrite (right)
##################### down-sampling #####################
N = 2 # down sampling 2 folds
intra_trace,t_d = resample(intra_trace, intra_trace.shape[0]/N, t)
extra_trace,t_d = resample(extra_trace, extra_trace.shape[0]/N, t)
t = t_d
print 'estimated fs ',1/(t[2]-t[1])
###################### filtering process #####################
fs = np.ceil(1/(t[2]-t[1]))
print('sampling frequency %f' % fs)
b, a = butter_bandpass(200,3000,fs,6)
extra_trace = filtfilt(b, a, extra_trace.T, padlen=150, padtype="even")
extra_trace = extra_trace.T
###############################################################
return _id, t, intra_trace, extra_trace, img_fname
elif i<0:
print('out of range, cannot less than 0')
elif i >= n_datum:
print('out of range, cannot less than %d' % n_datum)
def stochasticModel(x, w, N, stocf):
# x: input array sound, w: analysis window, N: FFT size,
# stocf: decimation factor of mag spectrum for stochastic analysis
# y: output sound
hN = N / 2 # size of positive spectrum
hM = (w.size) / 2 # half analysis window size
pin = hM # initialize sound pointer in middle of analysis window
fftbuffer = np.zeros(N) # initialize buffer for FFT
yw = np.zeros(w.size) # initialize output sound frame
w = w / sum(w) # normalize analysis window
ws = hanning(w.size) * 2 # synthesis window
# -----analysis-----
xw = x[pin - hM : pin + hM] * w # window the input sound
X = fft(xw) # compute FFT
mX = 20 * np.log10(abs(X[:hN])) # magnitude spectrum of positive frequencies
mXenv = resample(np.maximum(-200, mX), mX.size * stocf) # decimate the mag spectrum
pX = np.angle(X[:hN])
# -----synthesis-----
mY = resample(mXenv, hN) # interpolate to original size
pY = 2 * np.pi * np.random.rand(hN) # generate phase random values
Y = np.zeros(N, dtype=complex)
Y[:hN] = 10 ** (mY / 20) * np.exp(1j * pY) # generate positive freq.
Y[hN + 1 :] = 10 ** (mY[:0:-1] / 20) * np.exp(-1j * pY[:0:-1]) # generate negative freq.
fftbuffer = np.real(ifft(Y)) # inverse FFT
y = ws * fftbuffer * N / 2 # overlap-add
return mX, pX, mY, pY, y
def upsample_sig(self):
# Check that the upsample value is more than 0
# upsample the signal
self.cur_signal_up = signal.resample(self.cur_signal_raw, self.cur_signal_raw.size*self.N_up)
tmpr = signal.resample(np.real(self.cur_signal_raw), self.cur_signal_raw.size*self.N_up)
# change signal to magnitude of signal according to user specs
if self.sig_type == 'm':
self.cur_signal_up = self.sig_mag(self.cur_signal_up)
# if the user asked to filter, add observation to average
if self.filter_on:
self.filter_sig()
# if this is our first CIR, save it for reference, and flip flag
if self.is_first_obs:
self.start_time = self.cur_time
self.num_taps = self.cur_signal_raw.size
self.up_sig_len = self.cur_signal_up.size
self.ref_complex_cir = self.cur_signal_up.copy()
self.__init_mats()
self.is_first_obs = 0
def main():
X,Y = [],[]
Y2 = []
index = 0
with open(sys.argv[1],'r') as fp:
lines = [i.strip().split() for i in fp.readlines() if i]
for line in lines:
y = float(line[1])
x = index
index += 1
Y.append(y)
X.append(x)
if x > 1500 and x < 1700:
print x,y,y-y*3*math.exp(-(x-1600)**2/1500)
Y2.append(y-y*3*math.exp(-(x-1600)**2/1500))
else:
Y2.append(y)
y = signal.resample(Y,100)
y2 = signal.resample(Y2,100)
y = [i/max(y) for i in y]
y2 = [i/max(y2) for i in y2]
with open("%s.resampled" %sys.argv[1], 'w') as fp:
fp.write('x,y\n')
fp.write('\n'.join(["%s,%s" % (x,y) for x,y in zip(range(len(y)),y)]))
with open("%s.resampled.absorber" %sys.argv[1], 'w') as fp:
fp.write('x,y\n')
fp.write('\n'.join(["%s,%s" % (x,y) for x,y in zip(range(len(y2)),y2)]))
def stochasticModel(x, H, stocf): # stochastic analysis/synthesis of a sound, one frame at a time # x: input array sound, H: hop size, # stocf: decimation factor of mag spectrum for stochastic analysis # returns y: output sound N = H*2 # FFT size w = hanning(N) # analysis/synthesis window x = np.append(np.zeros(H),x) # add zeros at beginning to center first window at sample 0 x = np.append(x,np.zeros(H)) # add zeros at the end to analyze last sample pin = 0 # initialize sound pointer in middle of analysis window pend = x.size-N # last sample to start a frame y = np.zeros(x.size) # initialize output array while pin<=pend: #-----analysis----- xw = x[pin:pin+N]*w # window the input sound X = fft(xw) # compute FFT mX = 20 * np.log10(abs(X[:H])) # magnitude spectrum of positive frequencies mYst = resample(np.maximum(-200, mX), mX.size*stocf) # decimate the mag spectrum #-----synthesis----- mY = resample(mYst, H) # interpolate to original size pY = 2*np.pi*np.random.rand(H) # generate phase random values Y = np.zeros(N, dtype = complex) Y[:H] = 10**(mY/20) * np.exp(1j*pY) # generate positive freq. Y[H+1:] = 10**(mY[:0:-1]/20) * np.exp(-1j*pY[:0:-1]) # generate negative freq. fftbuffer = np.real(ifft(Y)) # inverse FFT y[pin:pin+N] += w*fftbuffer # overlap-add pin += H y = np.delete(y, range(H)) # delete half of first window which was added y = np.delete(y, range(y.size-H, y.size)) # delete half of last window which was added # advance sound pointer return y
def subsample_align_upsampling(self, sig_tx, sig_rx, n_up=32):
"""
Returns an aligned version of sig_tx and sig_rx by cropping and subsample alignment
Using upsampling
"""
assert (sig_tx.shape[0] == sig_rx.shape[0])
if sig_tx.shape[0] % 2 == 1:
sig_tx = sig_tx[:-1]
sig_rx = sig_rx[:-1]
sig1_up = signal.resample(sig_tx, sig_tx.shape[0] * n_up)
sig2_up = signal.resample(sig_rx, sig_rx.shape[0] * n_up)
off_meas = self.lag_upsampling(sig2_up, sig1_up, n_up=1)
off = int(abs(off_meas))
if off_meas > 0:
sig1_up = sig1_up[:-off]
sig2_up = sig2_up[off:]
elif off_meas < 0:
sig1_up = sig1_up[off:]
sig2_up = sig2_up[:-off]
sig_tx = signal.resample(sig1_up, sig1_up.shape[0] / n_up).astype(np.complex64)
sig_rx = signal.resample(sig2_up, sig2_up.shape[0] / n_up).astype(np.complex64)
return sig_tx, sig_rx
def data_parser(theta,kappa,tt,ch,tt_ch):
theta_r = np.array([[resample(theta.values.squeeze()[i,950:1440],50)] for i in range(0,theta.shape[0])])
theta_r = zscore(theta_r.squeeze(),axis=None)
kappa_r = np.array([[resample(kappa.values.squeeze()[i,950:1440],50)] for i in range(0,kappa.shape[0])])
kappa_r = zscore(kappa_r.squeeze(),axis=None)
kappa_df = pd.DataFrame(kappa_r)
theta_df = pd.DataFrame(theta_r)
both_df = pd.concat([theta_df,kappa_df],axis=1)
if tt_ch == 'tt':
# trial type
clean = np.nan_to_num(tt) !=0
tt_c = tt[clean.squeeze()].values
else :
# choice
clean = np.nan_to_num(ch) !=0
tt_c = ch[clean.squeeze()].values
# tt_c = tt[tt.values !=0|3].values
both = both_df.values
# both_c = both[clean.squeeze(),:]
both_c = both[clean.squeeze(),:]
# keeping one hot vector for now (incase we want it later)
# labs = np.eye(3)[tt_c.astype(int)-1]
# y[np.arange(3), a] = 1
# labs = labs.squeeze()
return both_c, tt_c, clean
def rx_oversampled(frames, ref_frame, modulated_frame, x_preamble, data, rx_kernel, demapper, timeslots, fft_len, cp_len, cs_len):
ref_frame_os = signal.resample(ref_frame, 2 * len(ref_frame))
x_preamble_os = signal.resample(x_preamble, 2 * len(x_preamble))
nyquist_frame_len = cp_len + 2 * fft_len + cs_len + cp_len + timeslots * fft_len + cs_len
n_frames = np.shape(frames)[0]
sync_frames = np.zeros((n_frames, nyquist_frame_len), dtype=np.complex)
print('nyquist sampled frame len', nyquist_frame_len, 'with n_frames', n_frames)
f_start = cp_len + 2 * fft_len + cs_len
d_start = f_start + cp_len
print('data start: ', d_start)
for i, f in enumerate(frames[0:2]):
tf = np.roll(f, 1)
tf[0] = 0
ff = signal.resample(tf, len(f) // 2)
sframe = synchronize_time(ff, ref_frame_os, x_preamble_os, 2 * fft_len, 2 * cp_len)
sframe = signal.resample(sframe, len(sframe) // 2)
sframe = synchronize_freq_offsets(sframe, modulated_frame, x_preamble, fft_len, cp_len, samp_rate=3.125e6)
print(len(sframe), len(ref_frame))
rx_preamble = sframe[cp_len:cp_len + 2 * fft_len]
avg_phase = calculate_avg_phase(rx_preamble, x_preamble)
# m, c = calculate_avg_phase(rx_preamble, x_preamble)
# avg_phase = calculate_avg_phase(sframe, ref_frame)
# phase_eqs = m * np.arange(-cp_len, len(sframe) - cp_len) + c
# sframe *= np.exp(-1j * phase_eqs)
# sframe *= np.exp(-1j * avg_phase)
sync_frames[i] = sframe
rx_data_frame = sframe[d_start:d_start + fft_len * timeslots]
# # rx_data_frame *= np.exp(-1j * avg_phase)
#
demodulate_frame(rx_data_frame, modulated_frame, rx_kernel, demapper, data, timeslots, fft_len)
for i, f in enumerate(sync_frames[0:3]):
rx_data_frame = f[d_start:d_start + fft_len * timeslots]
demodulate_frame(rx_data_frame, modulated_frame, rx_kernel, demapper, data, timeslots, fft_len)
def filter(self):
'''
resamples time series
:return:resampled time series with sampling frequency set to resamplerate
'''
# samplerate = self.time_series.attrs['samplerate']
samplerate = float(self.time_series['samplerate'])
time_axis_length = np.squeeze(self.time_series.coords['time'].shape)
new_length = int(np.round(time_axis_length*self.resamplerate/samplerate))
print new_length
if self.time_axis_index<0:
self.time_axis_index = self.time_series.get_axis_num('time')
time_axis = self.time_series.coords[ self.time_series.dims[self.time_axis_index] ]
try:
time_axis_data = time_axis.data['time'] # time axis can be recarray with one of the arrays being time
except (KeyError ,IndexError) as excp:
# if we get here then most likely time axis is ndarray of floats
time_axis_data = time_axis.data
time_idx_array = np.arange(len(time_axis))
if self.round_to_original_timepoints:
filtered_array, new_time_idx_array = resample(self.time_series.data,
new_length, t=time_idx_array,
axis=self.time_axis_index, window=self.window)
# print new_time_axis
new_time_idx_array = np.rint(new_time_idx_array).astype(np.int)
new_time_axis = time_axis[new_time_idx_array]
else:
filtered_array, new_time_axis = resample(self.time_series.data,
new_length, t=time_axis_data,
axis=self.time_axis_index, window=self.window)
coords = []
for i, dim_name in enumerate(self.time_series.dims):
if i != self.time_axis_index:
coords.append(self.time_series.coords[dim_name].copy())
else:
coords.append((dim_name,new_time_axis))
filtered_time_series = xray.DataArray(filtered_array, coords=coords)
# filtered_time_series.attrs['samplerate'] = self.resamplerate
filtered_time_series['samplerate'] = self.resamplerate
return TimeSeriesX(filtered_time_series)
def resample(self, times=2):
from scipy.signal import resample
self.flux, self.freq = resample(self.flux, int(len(self.flux)/times),
t=self.freq)
if hasattr(self, "fluxcal"):
self.fluxcal = resample(self.fluxcal,
int(len(self.fluxcal)/times))
if hasattr(self, "baseline"):
self.baseline = resample(self.baseline,
int(len(self.baseline)/times))
def lag_upsampling(self, sig_orig, sig_rec, n_up):
if n_up != 1:
sig_orig_up = signal.resample(sig_orig, sig_orig.shape[0] * n_up)
sig_rec_up = signal.resample(sig_rec, sig_rec.shape[0] * n_up)
else:
sig_orig_up = sig_orig
sig_rec_up = sig_rec
l = self.lag(sig_orig_up, sig_rec_up)
l_orig = float(l) / n_up
return l_orig
def RotateVector_45degree(newN,vec):
from scipy import ndimage as ndi
from scipy import signal
N = vec.shape[0]
newvec = numpy.zeros((newN,newN,3),float)
for i in range(3):
temp = signal.resample(signal.resample(vec[:,:,i],N*4,axis=0),N*4,axis=1)
temp = ndi.rotate(temp,45,mode='wrap')
newvec[:,:,i] = signal.resample(signal.resample(temp,newN,axis=0),newN,axis=1)
return newvec
def resample_via_fft(curve, samples):
""" resample_via_fft(curve, samples)
Resample the curve using scily signal processing utility via Fourier and zero padding
"""
rx = resample(curve[0,:], samples)
ry = resample(curve[1,:], samples)
# rsig = resample(curve, samples, axis=0)
return np.array([rx, ry])
def resample_with_gaussian_blur(input_array,sigma_for_gaussian,resampling_factor): sz=input_array.shape gauss_temp=ndimage.gaussian_filter(input_array,sigma=sigma_for_gaussian) resam_temp=sg.resample(gauss_temp,axis=1,num=sz[1]/resampling_factor) resam_temp=sg.resample(resam_temp,axis=2,num=sz[2]/resampling_factor) return (resam_temp)
def fft_gauss_blur_img(img, scale, std_cut_off=5):
old_img_size = img.shape[0]
new_img_size = np.round( img.shape[0]*scale )
std = guassianDownSampleSTD( old_img_size , new_img_size, std_cut_off, old_img_size )
sr = sig.resample( img, old_img_size, window = ('gaussian', std) )
sr = sig.resample( sr, old_img_size, window = ('gaussian', std), axis=1)
return sr
def fft_resample_img(img, nPix, std_cut_off = None):
#this is only for square images
if not std_cut_off is None:
oldSize = np.size(img,0)
std = guassianDownSampleSTD( oldSize, nPix, stdCutOff, oldSize )
sr = sig.resample(img, nPix, window = ('gaussian', std))
sr = sig.resample(sr, nPix, window = ('gaussian', std ), axis = 1)
else:
sr = sig.resample( img, nPix, window = ('boxcar') )
sr = sig.resample( sr, nPix, window = ('boxcar'), axis = 1)
return sr
def stftMorph(x1, x2, fs, w1, N1, w2, N2, H1, smoothf, balancef):
"""
Morph of two sounds using the STFT
x1, x2: input sounds, fs: sampling rate
w1, w2: analysis windows, N1, N2: FFT sizes, H1: hop size
smoothf: smooth factor of sound 2, bigger than 0 to max of 1, where 1 is no smothing,
balancef: balance between the 2 sounds, from 0 to 1, where 0 is sound 1 and 1 is sound 2
returns y: output sound
"""
if (N2/2*smoothf < 3): # raise exception if decimation factor too small
raise ValueError("Smooth factor too small")
if (smoothf > 1): # raise exception if decimation factor too big
raise ValueError("Smooth factor above 1")
if (balancef > 1 or balancef < 0): # raise exception if balancef outside 0-1
raise ValueError("Balance factor outside range")
if (H1 <= 0): # raise error if hop size 0 or negative
raise ValueError("Hop size (H1) smaller or equal to 0")
M1 = w1.size # size of analysis window
hM1_1 = int(math.floor((M1+1)/2)) # half analysis window size by rounding
hM1_2 = int(math.floor(M1/2)) # half analysis window size by floor
L = int(x1.size/H1) # number of frames for x1
x1 = np.append(np.zeros(hM1_2),x1) # add zeros at beginning to center first window at sample 0
x1 = np.append(x1,np.zeros(hM1_1)) # add zeros at the end to analyze last sample
pin1 = hM1_1 # initialize sound pointer in middle of analysis window
w1 = w1 / sum(w1) # normalize analysis window
M2 = w2.size # size of analysis window
hM2_1 = int(math.floor((M2+1)/2)) # half analysis window size by rounding
hM2_2 = int(math.floor(M2/2)) # half analysis window size by floor2
H2 = int(x2.size/L) # hop size for second sound
x2 = np.append(np.zeros(hM2_2),x2) # add zeros at beginning to center first window at sample 0
x2 = np.append(x2,np.zeros(hM2_1)) # add zeros at the end to analyze last sample
pin2 = hM2_1 # initialize sound pointer in middle of analysis window
y = np.zeros(x1.size) # initialize output array
for l in range(L):
#-----analysis-----
mX1, pX1 = DFT.dftAnal(x1[pin1-hM1_1:pin1+hM1_2], w1, N1) # compute dft
mX2, pX2 = DFT.dftAnal(x2[pin2-hM2_1:pin2+hM2_2], w2, N2) # compute dft
#-----transformation-----
mX2smooth = resample(np.maximum(-200, mX2), int(mX2.size*smoothf)) # smooth spectrum of second sound
mX2 = resample(mX2smooth, mX1.size) # generate back the same size spectrum
mY = balancef * mX2 + (1-balancef) * mX1 # generate output spectrum
#-----synthesis-----
y[pin1-hM1_1:pin1+hM1_2] += H1*DFT.dftSynth(mY, pX1, M1) # overlap-add to generate output sound
pin1 += H1 # advance sound pointer
pin2 += H2 # advance sound pointer
y = np.delete(y, range(hM1_2)) # delete half of first window which was added in stftAnal
y = np.delete(y, range(y.size-hM1_1, y.size)) # add zeros at the end to analyze last sample
return y
def resample(signal, prev_sample_rate, new_sample_rate):
if prev_sample_rate == new_sample_rate:
return signal
rate_factor = new_sample_rate/float(prev_sample_rate)
num_samples = int(len(signal.T)*rate_factor)
if signal.ndim == 2:
result = numpy.empty((len(signal), num_samples), dtype=signal.dtype)
for si, s in enumerate(signal):
result[si] = scisig.resample(s, num_samples)
return result
else:
return scisig.resample(signal, num_samples)
def _forward_dataset_helper(self, ds):
# local binding
num = self.__num
pos = None
if not self.__position_attr is None:
# we know something about sample position
pos = ds.sa[self.__position_attr].value
rsamples, pos = resample(ds.samples, self.__num, t=pos,
window=self.__window_args)
else:
# we know nothing about samples position
rsamples = resample(ds.samples, self.__num, t=None,
window=self.__window_args)
# new dataset that reuses that feature and dataset attributes of the
# source
mds = Dataset(rsamples, fa=ds.fa, a=ds.a)
# the tricky part is what to do with the samples attributes, since their
# number has changes
if self.__attr_strategy == 'remove':
# nothing to be done
pass
elif self.__attr_strategy == 'sample':
step = int(len(ds) / num)
sa = dict([(k, ds.sa[k].value[0::step][:num]) for k in ds.sa])
mds.sa.update(sa)
elif self.__attr_strategy == 'resample':
# resample the attributes themselves
sa = {}
for k in ds.sa:
v = ds.sa[k].value
if pos is None:
sa[k] = resample(v, self.__num, t=None,
window=self.__window_args)
else:
if k == self.__position_attr:
# position attr will be handled separately at the end
continue
sa[k] = resample(v, self.__num, t=pos,
window=self.__window_args)[0]
# inject them all
mds.sa.update(sa)
else:
raise ValueError("Unkown attribute handling strategy '%s'."
% self.__attr_strategy)
if not pos is None:
# we got the new sample positions and can store them
mds.sa[self.__position_attr] = pos
return mds
def resample(self, sampling_times):
[self.x, self.time] = signal.resample(
self.x,
len(sampling_times),
sampling_times)
[self.y, self.time] = signal.resample(
self.y,
len(sampling_times),
sampling_times)
[self.z, self.time] = signal.resample(
self.z,
len(sampling_times),
sampling_times)
def audio_loop(file, p, ratio, end, chunk_size, stop_proc):
time.sleep(0.5)
proc = psutil.Process(os.getpid())
proc.nice(-5)
time.sleep(0.02)
print ('audio file is ' + str(file))
while True:
#chunk = 2048/2
wf = wave.open(file, 'rb')
time.sleep(0.03)
data = wf.readframes(chunk_size.value)
time.sleep(0.03)
stream = p.open(
format = p.get_format_from_width(wf.getsampwidth()),
channels = wf.getnchannels(),
rate = wf.getframerate(),
output = True,
frames_per_buffer = chunk_size.value)
while data != '' and stop_proc.value == 0:
#need to try locking here for multiprocessing
array = numpy.fromstring(data, dtype=numpy.int16)
result = numpy.reshape(array, (array.size/2, 2))
#split data into seperate channels and resample
final = numpy.ones((1024,2))
reshapel = signal.resample(result[:, 0], 1024)
final[:, 0] = reshapel
reshaper = signal.resample(result[:, 1], 1024)
final[:, 1] = reshaper
out_data = final.flatten().astype(numpy.int16).tostring()
#data = signal.resample(array, chunk_size.value*ratio.value)
#stream.write(data.astype(int).tostring())
stream.write(out_data)
round_data = (int)(chunk_size.value*ratio.value)
if round_data % 2 != 0:
round_data += 1
data = wf.readframes(round_data)
stream.stop_stream()
stream.close()
wf.close()
p.terminate()
if end or stop_proc.value == 1:
break
stream.stop_stream()
stream.close()
wf.close()
p.terminate()
def resample(self, num, axis=0, window=None):
"""Resample timeseries data for a signal onto a different time
base. This routine uses scipy.signal's resample function which
creates an interpolation with the option of filtering by
specifying the window keyword. See help(scipy.signal.resample)
for details."""
if self['time'] is not None:
(self['signal'], self['time']) = resample(self['signal'], num,
t=self['time'], axis=0,
window=None)
else:
self['signal'] = resample(self['signal'], num, axis=0,
window=None)
def demodulate(sdr_samples, sample_rate, ignore=0):
t = np.arange(len(sdr_samples)) / sample_rate
demod = np.exp(-2j*pi*freq_offset*t)
y = sdr_samples * demod
threshold = 0.04
y = y[abs(y) > threshold]
from scipy import signal #should be imported already?
sig = angle(y[1:] * conj(y[:-1]))
h = signal.firwin(256,5000.0,nyq=sample_rate/2.0)
sigf = signal.fftconvolve(sig, h)
if ignore > 0:
sigf = sigf[ignore:-ignore]
downsample = 24
sigfd = sigf[::downsample]
fs_down = fs
sigfdd = signal.resample(sigfd, len(sigfd) * float(fs_down) / (sample_rate / float(downsample)))
return sigfdd
def resample(audio, target_sample_rate):
"""
Resamples audio to a specified sample rate.
This function should only be used for relatively short audio segments,
say not longer than a second or so. It uses the `scipy.signal.resample`
method to perform the resampling, which computes a length-M DFT and a
length-N inverse DFT, where M and N are the input and output length,
respectively. M and N may not be powers of two, and they may even be
prime, which can make this function slow if M or N is too large.
"""
if audio.sample_rate == target_sample_rate:
# do not need to resample
return audio
else:
# need to resample
# We put this import here instead of at the top of this module
# so the module can be used in Python environments that don't
# include SciPy as long as this function is not called.
import scipy.signal as signal
ratio = target_sample_rate / audio.sample_rate
num_samples = int(round(len(audio.samples) * ratio))
samples = signal.resample(audio.samples, num_samples)
return Bunch(samples=samples, sample_rate=target_sample_rate)
def stochasticModelSynth(stocEnv, H, N):
"""
Stochastic synthesis of a sound
stocEnv: stochastic envelope; H: hop size; N: fft size
returns y: output sound
"""
if not(UF.isPower2(N)): # raise error if N not a power of two
raise ValueError("N is not a power of two")
hN = N/2+1 # positive size of fft
No2 = N/2 # half of N
L = stocEnv[:,0].size # number of frames
ysize = H*(L+3) # output sound size
y = np.zeros(ysize) # initialize output array
ws = 2*hanning(N) # synthesis window
pout = 0 # output sound pointer
for l in range(L):
mY = resample(stocEnv[l,:], hN) # interpolate to original size
pY = 2*np.pi*np.random.rand(hN) # generate phase random values
Y = np.zeros(N, dtype = complex) # initialize synthesis spectrum
Y[:hN] = 10**(mY/20) * np.exp(1j*pY) # generate positive freq.
Y[hN:] = 10**(mY[-2:0:-1]/20) * np.exp(-1j*pY[-2:0:-1]) # generate negative freq.
fftbuffer = np.real(ifft(Y)) # inverse FFT
y[pout:pout+N] += ws*fftbuffer # overlap-add
pout += H
y = np.delete(y, range(No2)) # delete half of first window
y = np.delete(y, range(y.size-No2, y.size)) # delete half of the last window
return y
def convolve_scalogram(ana, wf, sampling_rate,optimize_fft):
n = wf.shape[0]
sig = ana.magnitude
ana_sr=ana.sampling_rate.rescale('Hz').magnitude
if optimize_fft:
sig=sig-sig.mean() # Remove mean before padding
nfft=int(2**np.ceil(np.log(sig.size)/np.log(2)))
sig=np.r_[sig,np.zeros(nfft-sig.size)] # pad signal with 0 to a power of 2 length
sig=resample(sig,int(sig.size*sampling_rate/ana_sr)) # resample in time domain
sigf=fftpack.fft(sig,n) # Compute fft with a power of 2 length
else:
sigf=fftpack.fft(sig)
# subsampling in fft domain (attention factor)
factor = (sampling_rate/ana.sampling_rate).simplified.magnitude
x=(n-1)//2
if np.mod(n,2)==0:
sigf = np.concatenate([sigf[0:x+2], sigf[-x:]])*factor
else:
sigf = np.concatenate([sigf[0:x+1], sigf[-x:]])*factor
# windowing ???
#win = fftpack.ifftshift(np.hamming(n))
#sigf *= win
# Convolve (mult. in Fourier space)
wt_tmp=fftpack.ifft(sigf[:,np.newaxis]*wf,axis=0)
# and shift
wt = fftpack.fftshift(wt_tmp,axes=[0])
return wt
def stochasticResidualAnal(x, N, H, sfreq, smag, sphase, fs, stocf): """ Subtract sinusoids from a sound and approximate the residual with an envelope x: input sound, N: fft size, H: hop-size sfreq, smag, sphase: sinusoidal frequencies, magnitudes and phases fs: sampling rate; stocf: stochastic factor, used in the approximation returns stocEnv: stochastic approximation of residual """ hN = N/2 # half of fft size x = np.append(np.zeros(hN),x) # add zeros at beginning to center first window at sample 0 x = np.append(x,np.zeros(hN)) # add zeros at the end to analyze last sample bh = blackmanharris(N) # synthesis window w = bh/ sum(bh) # normalize synthesis window L = sfreq.shape[0] # number of frames, this works if no sines pin = 0 for l in range(L): xw = x[pin:pin+N] * w # window the input sound X = fft(fftshift(xw)) # compute FFT Yh = UF_C.genSpecSines(N*sfreq[l,:]/fs, smag[l,:], sphase[l,:], N) # generate spec sines Xr = X-Yh # subtract sines from original spectrum mXr = 20*np.log10(abs(Xr[:hN])) # magnitude spectrum of residual mXrenv = resample(np.maximum(-200, mXr), mXr.size*stocf) # decimate the mag spectrum if l == 0: # if first frame stocEnv = np.array([mXrenv]) else: # rest of frames stocEnv = np.vstack((stocEnv, np.array([mXrenv]))) pin += H # advance sound pointer return stocEnv
def make_step(self, size_in, size_out, dt, rng):
assert size_in[0] == 0
assert size_out[0] == 1
rate = 1. / dt
orig_rate, orig = readwav(self.path)
new_size = int(orig.size * (rate / orig_rate))
wave = resample(orig, new_size)
wave -= wave.mean()
# Normalize wave to desired rms
wave_rms = npext.rms(wave)
wave *= (self.rms / wave_rms)
if self.at_end == 'loop':
def step_wavfileloop(t):
idx = int(t * rate) % wave.size
return wave[idx]
return step_wavfileloop
elif self.at_end == 'stop':
def step_wavfilestop(t):
idx = int(t * rate)
if idx > wave.size:
return 0.
else:
return wave[idx]
return step_wavfilestop
def to_envelopes(path,num_bands,freq_lims,window_length=None,time_step=None):
sr, proc = preproc(path,alpha=0.97)
proc = proc/sqrt(mean(proc**2))*0.03;
bandLo = [ freq_lims[0]*exp(log(freq_lims[1]/freq_lims[0])/num_bands)**x for x in range(num_bands)]
bandHi = [ freq_lims[0]*exp(log(freq_lims[1]/freq_lims[0])/num_bands)**(x+1) for x in range(num_bands)]
if window_length is not None and time_step is not None:
use_windows = True
nperseg = int(window_length*sr)
noverlap = int(time_step*sr)
window = hanning(nperseg+2)[1:nperseg+1]
step = nperseg - noverlap
indices = arange(0, proc.shape[-1]-nperseg+1, step)
num_frames = len(indices)
envelopes = zeros((num_bands,num_frames))
else:
use_windows=False
sr_env = 120
t = len(proc)/sr
numsamp = ceil(t * sr_env)
envelopes = []
for i in range(num_bands):
b, a = butter(2,(bandLo[i]/(sr/2),bandHi[i]/(sr/2)), btype = 'bandpass')
env = filtfilt(b,a,proc)
env = abs(hilbert(env))
if use_windows:
window_sums = []
for k,ind in enumerate(indices):
seg = env[ind:ind+nperseg] * window
window_sums.append(sum(seg))
envelopes[i,:] = window_sums
else:
env = resample(env,numsamp)
envelopes.append(env)
return array(envelopes).T
def ups(array):
y = sg.resample(array, 600)
y = pd.DataFrame(y)
return (y)
#else:
# transforms[i] = [0,0,0]
prev_gray = curr_gray
#plt.plot(np.arange(transforms.shape[0]) * 1/29.97,transforms[:,0])
#plt.plot(np.arange(transforms.shape[0]) * 1/29.97,transforms[:,1])
#plt.plot(np.arange(transforms.shape[0]) * 1/29.97,transforms[:,2])
transforms = np.genfromtxt('test_clips/GX016017.MP4' + "opticalflowH6.csv",
delimiter=',')
mysample = transforms[0:100, 2] * 59.94
N = 0
resampled = resample(mysample, 400)
sampleB = integrator.get_raw_data("z")[N:400 + N]
testcorr = np.correlate(sampleB, resampled, "full")
plt.plot(testcorr)
print(np.argmax(testcorr) - 400)
maxcorr = np.argmax(testcorr) - 400
#plt.plot(resampled)
#plt.plot(integrator.get_raw_data("z")[0:400])
plt.plot(np.arange(mysample.shape[0]) * interval - 3.1 / 60, mysample)
#plt.figure()
counter=counter+1
yClass=[]
for i in range(0,len(y)):
temp=[0] * 11
temp[y[i]]=1
yClass.append(temp)
Xnew=[]
counter=0
for i in X:
f = signal.resample(i, 8000)
counter=1+counter
Xnew.append(f)
Xtest= np.array(Xnew)
ytest= np.array(yClass)
b = Xtest[:, :, newaxis]
Xtest=b
model = tf.keras.models.load_model('model1.h5') #Model name here.
pred= model.predict(Xtest,verbose=1)
#########
def GetAPTs(filename,
ti1=0,
ti2=1,
pval=100,
pad=False,
interpol=False,
norm=False):
print('Generating Pulse Train From File:')
print(filename)
print('\n')
## Spectral arrays.
Specx = np.zeros(shape=2048, dtype=complex)
Specy = np.zeros(shape=2048, dtype=complex)
ndt = 2048
# Unpack the datas.
Horder, xp, ixp, yp, iyp = np.genfromtxt(filename, unpack=True)
# Get x and y spectral components.
specx = xp + 1j * ixp
specy = yp + 1j * iyp
# Shift the spectra.
Specx[:1024] = specx
Specy[:1024] = specy
for idx, energy in enumerate(Horder):
if energy < 12.0:
Specx[idx] = 0
Specy[idx] = 0
if energy > 45:
Specx[idx] = 0
Specy[idx] = 0
if pad == True:
# Try zero padding hard core before ifft.
Specx = np.pad(Specx, (pval, pval), 'constant', constant_values=(0, 0))
Specy = np.pad(Specy, (pval, pval), 'constant', constant_values=(0, 0))
# Get the padded fields.
Ex = np.fft.ifft(Specx)
Ey = np.fft.ifft(Specy)
else:
# Get the fields.
Ex = np.fft.ifft(Specx)
Ey = np.fft.ifft(Specy)
# Shift the fields to center them at t=0.
fl = len(Ex) / 2
Ex = np.hstack((Ex[-fl:], Ex[:-fl]))
Ey = np.hstack((Ey[-fl:], Ey[:-fl]))
# Now, get the real part of the fields.
Ex = np.real(Ex)
Ey = np.real(Ey)
if interpol == True:
# Try interpolating.
Ex = signal.resample(Ex, len(Ex) * 3)
Ey = signal.resample(Ey, len(Ey) * 3)
# Create the time array.
dw = Horder[2] - Horder[1] # Frequency step size.
wmax = (Horder[-1] + dw) * omega # Max frequency.
t = np.linspace(0, len(Ex), len(Ex)) * np.pi / wmax # Time array.
dt = t[2] - t[1]
tmax = t[-1] + dt
tplot = (t - tmax / 2) / t_period
if norm == True:
Ex = Ex / np.max(np.abs(Ex))
Ey = Ey / np.max(np.abs(Ey))
# for idx, stuff in enumerate(Ex):
# print(idx, tplot[idx], np.sqrt(stuff**2 + Ey[idx]**2))
# Truncate arrays to a single cycle of the IR field.
tind = ((tplot >= ti1) & (tplot <= ti2))
tplot = tplot[tind]
Ex = Ex[tind]
Ey = Ey[tind]
# # Try padding the arrays for asthetics.
# tpre = np.linspace(tplot[0], tplot[0]-0.25, 50)
# tpost = np.linspace(tplot[-1], tplot[-1]+0.25, 50)
# tplot = np.concatenate((tpre, tplot, tpost))
# Ex = np.pad(Ex, (50, 50), 'constant', constant_values=(Ex[0], Ex[-1]))
# Ey = np.pad(Ey, (50, 50), 'constant', constant_values=(Ey[0], Ey[-1]))
return tplot, Ex, Ey
def convert(files, ids, genders, ecg_channel_name, dataset_name, i_patient):
"""
Extracts segments from the ECG recordings and saves them as .npy files
The segments of every patient will be stored in a separate folder.
args:
files (list of str): list of paths to the .edf recordings
ids (pd.Dataframe): internal patient ids
genders (pd.Dataframe): gender of each patient
ecg_channel_name (str) name of the channel with ECG-data in the .edf files
dataset_name (str): name of dataset ('shhs' or 'mesa')
i_patient (int): counter
"""
for file in files:
i_patient += 1
# Load index of file in meta-data
if dataset_name == 'shhs':
meta_idx = int(
np.argwhere(ids.values == int(file.split('-')[1][:-4])))
elif dataset_name == 'mesa':
meta_idx = int(
np.argwhere(ids.values == int(file.split('-')[-1][:-4])))
else:
raise Exception('Invalid dataset name. Select "shhs" or "mesa".')
# Get gender of patient from meta-data
gender = GENDER_DICT[genders.values[meta_idx]]
patient_id = str(ids.values[meta_idx]).zfill(6)
# Create new folder for the patient
patient_folder = os.path.join(
target_folder, '{}-{}-{}'.format(dataset_name, patient_id, gender))
os.makedirs(patient_folder, exist_ok=True)
# Load .edf file
with pyedflib.EdfReader(file) as f:
n = f.signals_in_file
signal_labels = f.getSignalLabels()
# Select ECG channel
for i_channel in range(n):
if signal_labels[i_channel] == ecg_channel_name:
channel = i_channel
# Load ECG channel
signal = f.readSignal(channel)
# Get sample frequency
fs = f.getSampleFrequency(channel)
# Set start 60 Minutes in the recording to avoid noise in the beginning
start = 60 * 60 * fs
# use only inner 5 hours where signal is cleaner
stop = min(
len(signal) - seg_len * fs - 60 * 60 * fs, 6 * 60 * 60 * fs)
if stop < start:
print("WARNING: stop < start for file {}".format(file))
continue
seg_indices = np.linspace(start,
stop,
num=config['num_segs_per_patient'],
dtype=np.int)
for i_segment, ind in enumerate(seg_indices):
local_idx = str(i_segment).zfill(2)
target_path = os.path.join(patient_folder,
'{}.npy'.format(local_idx))
segment = signal[ind:ind + seg_len * fs]
if plot_output:
plt.title("{}/{}, time: {:.2f}h/{:.2f}h".format(
i_patient, n_files_total, ind / (fs * 3600),
len(signal) / (fs * 3600)))
plt.plot(segment)
plt.show()
# Resample
if fs != config['new_fs']:
if abs(fs - config['new_fs']) > 20:
print(
"Warning, this is a big resampling, fs: {}, source: {}, target: {}"
.format(fs, file, target_path))
num = int(seg_len * new_fs)
segment = resample(segment, num=num)
# Save
if save:
np.save(file=target_path, arr=segment)
# Logging
if i_patient % config['log_interval'] == 0:
time_left = (((time.time() - t_start) / i_patient) *
(n_files_total - i_patient)) / 3600
print("\nProcessed file {} of {}".format(
i_patient, n_files_total))
print(target_path, fs, len(segment),
"{}/{}".format(i_patient, n_files_total),
"\nTime elapsed {:.2f}s".format(time.time() - t_start),
"Estimated time left: {:.2f} hours".format(time_left))
return i_patient
def preprocess(data, factor, low_perc=5, up_perc=80):
'''
Preprocess gcamp data based on https://github.com/lucastheis/c2s/blob/master/c2s/c2s.py#L166
Main differences:
- Simple linear regression to remove trends
- Allows standardization to differ in lower and upper percentiles
- Additional tools for managing camera and spike times
:param data: dict containing
'calcium' input gcamp (required) [n]
'fps' original fps (required) [1]
'cam_times' original camera times (optional) [n]
'spike_times' spike times in seconds (optional) [m]
:param factor: factor to upsample data. will produce upsampled data of size [factor * n]
:param low_perc: lower percentile to normalize dff
:param up_perc: upper percentile to normalize dff
:return: dict with upsampled data and binned spikes (if spike_times provided)
'''
if 'cam_times' in data:
assert data['calcium'].shape[0] == data['cam_times'].shape[
0], "gcamp and camera times must be same length"
data = deepcopy(data)
x = np.arange(data['calcium'].shape[0])
regr = linear_model.LinearRegression()
regr.fit(x.reshape(-1, 1), data['calcium'])
a = regr.coef_
b = regr.intercept_
data['calcium'] = data['calcium'] - (a * x + b)
calcium_low_perc = np.percentile(data['calcium'], low_perc)
calcium_high_perc = np.percentile(data['calcium'], up_perc)
if calcium_high_perc - calcium_low_perc > 0.:
data['calcium'] = (data['calcium'] - calcium_low_perc
) / float(calcium_high_perc - calcium_low_perc)
# normalize sampling rate
if factor > 1:
# number of samples after update of sampling rate
num_samples = data['calcium'].shape[0] * factor
# resample calcium signal
data['calcium'] = resample(data['calcium'].ravel(), num_samples)
data['fps'] = data['fps'] * factor
if 'cam_times' in data:
data['cam_times'] = time_adjustment(data['cam_times'], factor,
data['fps'])
else:
data['cam_times'] = np.arange(
0, data['calcium'].shape[0] / data['fps'], (1.0 / data['fps']))
data['cam_times'] = data['cam_times'][np.arange(
data['calcium'].shape[0])]
if 'spike_times' in data:
data['spikes_per_bin'] = bin_spikes(data['cam_times'],
data['spike_times'])
return data
def samples_at(self, rate):
new_samples = np.frombuffer(self.bytes_chunks, dtype=np.float32)
return signal.resample(new_samples, rate * self.duration)
def my_resample(x, origin_fs, target_fs):
from scipy.signal import resample
origin_l = len(x)
target_l = np.int32((1. * origin_l / origin_fs * target_fs))
xx = resample(x, target_l)
return xx
def resample(signal, fs, new_fs):
"""Resample `signal` from `fs` to `new_fs`."""
new_signal = dsp.resample(signal, int(len(signal) * new_fs / fs))
return new_signal
def __resize_matrix_resample(self, groupsrc, groupdst, M):
from scipy.signal import resample
x_resampled = resample(M, len(groupsrc), window='blk')
xy_resampled = resample(
x_resampled.transpose(), len(groupdst), window='blk').transpose()
return np.minimum(np.maximum(xy_resampled, 0), 1)
def main():
# Args
parser = argparse.ArgumentParser()
parser.add_argument("--model_name",
default='models/XXL_lr000001_bs32.pth',
help="Classifier model path")
parser.add_argument(
"--classifier",
default='XXL',
help="Choose classifier architecture, C, S, XS, XL, XXL, XXXL")
args = parser.parse_args()
# Select training device
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Load specified Classifier
if args.classifier == 'XS':
net = Classifier_XS()
elif args.classifier == 'S':
net = Classifier_S()
elif args.classifier == 'XL':
net = Classifier_XL()
elif args.classifier == 'XXL':
net = Classifier_XXL()
elif args.classifier == 'XXXL':
net = Classifier_XXXL()
else:
net = Classifier()
net.to(device)
# Load parameters from trained model
net.load_state_dict(
torch.load('../../STEAD_ANN/models/' + args.model_name + '.pth'))
net.eval()
# Load Belgica data
f = scipy.io.loadmat(
"../../Data_Belgica/mat_2018_08_19_00h28m05s_Parkwind_HDAS_2Dmap_StrainData_2D.mat"
)
# Read data
data = f['Data_2D']
fs = 10
total = 0
seis, ns = 0, 0
# For every trace in the file
for trace in data:
# Resample
resamp_trace = signal.resample(trace, 6000)
# Normalize
resamp_trace = resamp_trace / np.max(np.abs(resamp_trace))
# Numpy to Torch
resamp_trace = torch.from_numpy(resamp_trace).to(device).unsqueeze(0)
# Prediction
outputs = net(resamp_trace.float())
predicted = torch.round(outputs.data).item()
# Count traces
total += 1
if predicted:
seis += 1
else:
ns += 1
# Average 5km of measurements
avg_data = np.mean(data[3500:4001, :], 0)
# Filter average data
avg_data_filtered1 = butter_bandpass_filter(avg_data, 0.5, 1, fs, order=5)
avg_data_filtered2 = butter_bandpass_filter(avg_data,
0.2,
0.6,
fs,
order=5)
avg_data_filtered3 = butter_bandpass_filter(avg_data,
0.1,
0.3,
fs,
order=5)
# Resample
avg_data = signal.resample(avg_data, 6000)
avg_data_filtered1 = signal.resample(avg_data_filtered1, 6000)
avg_data_filtered2 = signal.resample(avg_data_filtered2, 6000)
avg_data_filtered3 = signal.resample(avg_data_filtered3, 6000)
# Normalize
avg_data = avg_data / np.max(np.abs(avg_data))
avg_data_filtered1 = avg_data_filtered1 / np.max(
np.abs(avg_data_filtered1))
avg_data_filtered2 = avg_data_filtered2 / np.max(
np.abs(avg_data_filtered2))
avg_data_filtered3 = avg_data_filtered3 / np.max(
np.abs(avg_data_filtered3))
# Numpy to Torch
avg_data = torch.from_numpy(avg_data).to(device).unsqueeze(0)
avg_data_filtered1 = torch.from_numpy(avg_data_filtered1).to(
device).unsqueeze(0)
avg_data_filtered2 = torch.from_numpy(avg_data_filtered2).to(
device).unsqueeze(0)
avg_data_filtered3 = torch.from_numpy(avg_data_filtered3).to(
device).unsqueeze(0)
# Prediction
output = net(avg_data.float())
output_filtered1 = net(avg_data_filtered1.float())
output_filtered2 = net(avg_data_filtered2.float())
output_filtered3 = net(avg_data_filtered3.float())
predicted = torch.round(output.data).item()
predicted_filtered1 = torch.round(output_filtered1.data).item()
predicted_filtered2 = torch.round(output_filtered2.data).item()
predicted_filtered3 = torch.round(output_filtered3.data).item()
# Results
print(f'Inferencia Belgica:\n\n'
f'Total traces: {total}\n'
f'Total predicted seismic traces: {seis}\n'
f'Total predicted noise traces: {ns}\n')
print(f'Average predicted: {predicted}\n'
f'Average filtered predicted 1: {predicted_filtered1}\n'
f'Average filtered predicted 2: {predicted_filtered2}\n'
f'Average filtered predicted 3: {predicted_filtered3}\n')
data = pd.concat([train, test], sort=False)
sub = pd.read_csv('data/提交结果示例.csv')
y = train.groupby('fragment_id')['behavior_id'].min()
train['mod'] = (train.acc_x**2 + train.acc_y**2 + train.acc_z**2)**.5
train['modg'] = (train.acc_xg**2 + train.acc_yg**2 + train.acc_zg**2)**.5
test['mod'] = (test.acc_x**2 + test.acc_y**2 + test.acc_z**2)**.5
test['modg'] = (test.acc_xg**2 + test.acc_yg**2 + test.acc_zg**2)**.5
x = np.zeros((7292, 60, 8, 1))
t = np.zeros((7500, 60, 8, 1))
for i in tqdm(range(7292)):
tmp = train[train.fragment_id == i][:60]
x[i, :, :, 0] = resample(
tmp.drop(['fragment_id', 'time_point', 'behavior_id'], axis=1), 60,
np.array(tmp.time_point))[0]
for i in tqdm(range(7500)):
tmp = test[test.fragment_id == i][:60]
t[i, :, :, 0] = resample(tmp.drop(['fragment_id', 'time_point'], axis=1),
60, np.array(tmp.time_point))[0]
kfold = StratifiedKFold(5, shuffle=True)
def Net():
emb_size = 6
input = Input(shape=(60, 8, 1))
input_1_1 = Lambda(lambda x: x[:, :, 1:, :])(input)
input_1_2 = Lambda(lambda x: x[:, :, 0, :])(input)
def _resample_fs(cls, data, new_fs, old_fs):
fs_ratio = new_fs / old_fs
new_length = int(np.round(len(data) * fs_ratio))
return signal.resample(data, new_length)
def resample(data, fs, new_fs=44150):
resampled = signal.resample(data, int(len(data)*new_fs/fs))
return resampled
def proc_file(file):
fobj = bu.hsDat(file, load=True)
vperp = fobj.dat[:, 0]
elec3 = fobj.dat[:, 1]
try:
phi_dg = fobj.attribs['phi_dg']
except:
phi_dg = 0.0
inds = np.abs(full_freqs - fspin) < 200.0
cut = int(0.1 * fsamp)
zeros = np.zeros_like(elec3[:cut])
voltages = [
zeros, zeros, zeros, elec3[:cut], -1.0 * elec3[:cut], zeros, zeros,
zeros
]
efield = bu.trap_efield(voltages, only_x=True)[0]
drive_amp, drive_phase = bu.get_sine_amp_phase(efield)
elec3_fft = np.fft.rfft(elec3)
true_fspin = np.average(full_freqs[inds], weights=np.abs(elec3_fft)[inds])
carrier_amp, carrier_phase_mod = \
bu.demod(vperp, true_fspin, fsamp, plot=plot_carrier_demod, \
filt=True, bandwidth=bandwidth,
notch_freqs=notch_freqs, notch_qs=notch_qs, \
tukey=True, tukey_alpha=5.0e-4, \
detrend=True, detrend_order=1, harmind=2.0)
# b1, a1 = signal.butter(3, np.array(libration_filt_band)*2.0/fsamp, btype='bandpass')
sos = signal.butter(3,
libration_filt_band,
btype='bandpass',
fs=fsamp,
output='sos')
# carrier_phase_mod_filt = signal.filtfilt(b1, a1, carrier_phase_mod)
carrier_phase_mod_filt = signal.sosfiltfilt(sos, carrier_phase_mod)
if len(libration_filt_band):
libration_inds = (full_freqs > libration_filt_band[0]) \
* (full_freqs < libration_filt_band[1])
else:
libration_inds = np.abs(full_freqs -
libration_guess) < 0.5 * libration_bandwidth
phase_mod_fft = np.fft.rfft(carrier_phase_mod) * fac
lib_fit_x = full_freqs[libration_inds]
lib_fit_y = np.abs(phase_mod_fft[libration_inds])
try:
try:
peaks = bu.find_fft_peaks(lib_fit_x,
lib_fit_y,
delta_fac=5.0,
window=50)
ind = np.argmax(peaks[:, 1])
except:
peaks = bu.find_fft_peaks(lib_fit_x,
lib_fit_y,
delta_fac=3.0,
window=100)
ind = np.argmax(peaks[:, 1])
true_libration_freq = peaks[ind, 0]
except:
true_libration_freq = lib_fit_x[np.argmax(lib_fit_y)]
libration_amp, libration_phase = \
bu.demod(carrier_phase_mod, true_libration_freq, fsamp, \
plot=plot_libration_demod, filt=True, \
filt_band=libration_filt_band, \
bandwidth=libration_bandwidth, \
tukey=False, tukey_alpha=5.0e-4, \
detrend=False, detrend_order=1.0, harmind=1.0)
libration_ds, time_vec_ds = \
signal.resample(carrier_phase_mod_filt, t=time_vec, num=out_nsamp)
libration_amp_ds, time_vec_ds = \
signal.resample(libration_amp, t=time_vec, num=out_nsamp)
libration_ds = libration_ds[out_cut:int(-1 * out_cut)]
libration_amp_ds = libration_amp_ds[out_cut:int(-1 * out_cut)]
time_vec_ds = time_vec_ds[out_cut:int(-1 * out_cut)]
if plot_downsample:
plt.plot(time_vec,
carrier_phase_mod_filt,
color='C0',
label='Original')
plt.plot(time_vec_ds,
libration_ds,
color='C0',
ls='--',
label='Downsampled')
plt.plot(time_vec, libration_amp, color='C1') #, label='Original')
plt.plot(time_vec_ds, libration_amp_ds, color='C1',
ls='--') #, label='Downsampled')
plt.legend()
plt.show()
input()
return (time_vec_ds, libration_ds, libration_amp_ds, \
true_libration_freq, phi_dg, drive_amp)
def main():
# Args
parser = argparse.ArgumentParser()
parser.add_argument("--model_name", default='Default_model', help="Classifier model path")
parser.add_argument("--classifier", default='C', help="Choose classifier architecture, C, CBN")
args = parser.parse_args()
# Select training device
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Load specified Classifier
if args.classifier == 'CBN':
net = ClassConvBN()
elif args.classifier == 'C':
net = ClassConv()
else:
net = ClassConv()
print('Bad Classifier option, running classifier C')
net.to(device)
# Load parameters from trained model
net.load_state_dict(torch.load('../../STEAD_CNN/models/' + args.model_name + '.pth'))
net.eval()
# Load Francia data file 1
f = scipy.io.loadmat("../../Data_Francia/Earthquake_1p9_Var_BP_2p5_15Hz.mat")
# Read data
data = f["StrainFilt"]
# time= f["Time"]
# distance = f["Distance_fiber"]
# Sampling frequency
fs = 100
total = 0
tr_seismic, tr_noise = 0, 0
fil_seismic, fil_noise = 0, 0
# For every trace in the file
for trace in data:
if np.max(np.abs(trace)):
# Filter
fil_trace = butter_bandpass_filter(trace, 0.5, 1, fs, order=5)
# Resample
resamp_trace = signal.resample(trace, 6000)
resamp_fil_trace = signal.resample(fil_trace, 6000)
# Normalize
resamp_trace = resamp_trace / np.max(np.abs(resamp_trace))
resamp_fil_trace = resamp_fil_trace / np.max(np.abs(resamp_fil_trace))
# Numpy to Torch
resamp_trace = torch.from_numpy(resamp_trace).to(device).unsqueeze(0)
resamp_fil_trace = torch.from_numpy(resamp_fil_trace).to(device).unsqueeze(0)
# Prediction
out_trace = net(resamp_trace.float())
out_fil_trace = net(resamp_fil_trace.float())
pred_trace = torch.round(out_trace.data).item()
pred_fil_trace = torch.round(out_fil_trace.data).item()
# Count traces
total += 1
if pred_trace:
tr_seismic += 1
else:
tr_noise += 1
if pred_fil_trace:
fil_seismic += 1
else:
fil_noise += 1
# Results
print(f'Inferencia Francia:\n\n'
f'Total traces: {total}\n'
f'Predicted seismic: {tr_seismic}, predicted noise: {tr_noise}\n'
f'Predicted fil_seismic: {fil_seismic}, predicted fil_noise: {fil_noise}\n')
def feature_extractor(pair_of_attrs, pair_of_attr_names):
attr1 = pair_of_attrs[0]
# attr2 = pair_of_attrs[1]
attr1_name = pair_of_attr_names[0]
# attr2_name = pair_of_attr_names[1]
attr1_df = pd.DataFrame({attr1_name: attr1})
# attr2_df = pd.DataFrame({attr2_name: attr2})
content_ratio1 = content_ratio(attr1)
# content_ratio2 = content_ratio(attr2)
# print(content_ratio1)
# print(content_ratio2)
# all attrs in table1
tbl1_attrs = attr1_df.keys()
# all attrs in table2
# tbl2_attrs = attr2_df.keys()
distributions1 = {}
# distributions2 = {}
for attr in tbl1_attrs:
content_histogram_tbl1 = content_histogram(attr1_df, attr)
distributions1[attr] = content_histogram_tbl1
# for attr in tbl2_attrs:
# content_histogram_tbl2 = content_histogram(attr2_df, attr)
# distributions2[attr] = content_histogram_tbl2
# table1
attr1_histog = []
for distr in distributions1:
# print(distr)
if distr == attr1_name:
# distributions1[distr].sort_values(distr)
for index, row in distributions1[distr].iterrows():
# print(row)
# for key in row.keys():
attr1_histog.append(row[distr])
# # table2
# attr2_histog = []
# for distr in distributions2:
# # print(distr)
# if distr == attr2_name:
# # distributions2[distr].sort_values(distr)
# for index, row in distributions2[distr].iterrows():
# attr2_histog.append(row[distr])
attr1_histog.sort()
# attr2_histog.sort()
# print(attr1_histog)
# print(attr2_histog)
# resample using fft
resample_dimension = 20
attr1_histog_re = signal.resample(attr1_histog, resample_dimension)
# attr2_histog_re = signal.resample(attr2_histog, resample_dimension)
# print(attr1_histog_re)
# # print(attr2_histog_re)
# visualize_histograms_pair(attr1_name, attr1_histog_re, attr2_name, attr2_histog_re, resample_dimension)
# print(_get_col_dtype(attr1_df[attr1_name]))
# print(_get_col_dtype(attr2_df[attr2_name]))
average_cell_len1 = average_cell_len(attr1)
# average_cell_len2 = average_cell_len(attr2)
percentage_of_num1, percentage_of_alphabetic1 = percentage_of_num_alphabetic(
attr1)
# percentage_of_num2, percentage_of_alphabetic2 = percentage_of_num_alphabetic(attr2)
attr1_features = [content_ratio1]
attr1_features.extend(attr1_histog_re)
attr1_features.append(average_cell_len1)
attr1_features.append(percentage_of_num1)
attr1_features.append(percentage_of_alphabetic1)
# attr2_features = [content_ratio2]
# attr2_features.extend(attr2_histog_re)
# attr2_features.append(average_cell_len2)
# attr2_features.append(percentage_of_num2)
# attr2_features.append(percentage_of_alphabetic2)
attr1_features_dict = ['attr_histogram'] * resample_dimension
attr1_features_dict = [
'content_ratio', *attr1_features_dict, 'average_cell_len',
'percentage_of_num', 'percentage_of_alphabetic'
]
return attr1_features, attr1_features_dict
import numpy as np
import matplotlib.pyplot as plot
import scipy.signal as sg
import wave
from scipy.io.wavfile import read
rate, signal = read("../Xe.wav")
maxIndex = len(signal)
plot.figure()
for frames in range(0, 20):
signalArray = []
signalTQ = []
for k in range(6000 + frames * 256, 6300 + frames * 256):
signalArray.append(signal[k])
signalArray = sg.resample(signalArray, 300)
for k in range(0, 149): # day la K 0..150
sig = 0
for y in range(0, 299 - k):
sig = sig + signalArray[y] * signalArray[y + k] # day la R(k)
signalTQ.append(sig)
plot.figure(1)
plot.clf()
# plot.subplot(111)
plot.plot(signalTQ, color="purple")
plot.grid(True)
plot.figure(2)
plot.clf()
plot.subplot(311)
def get_data(self): # Function to get force data from txt file
filename = input(
"Enter Force File Name with full Path if not in Working Directory (Ex: User/File.txt): "
)
print(" ")
file1 = open(filename) # set file name
f1 = file1.readlines() # read file
DataBegin = 19 # Begin at line 19 to get force data
for i in range(DataBegin, len(f1)):
cnt = 0
for x in range(len(f1[i])):
if (f1[i][x] == '\t'):
if (cnt == 0):
self.time_abs.append(float(
f1[i][0:x])) # Add to time list
temp = x + 1
if (cnt == 1):
self.forceX.append(float(
f1[i][temp:x])) # Add to Attila X list
temp = x + 1
if (cnt == 2):
self.forceY.append(float(
f1[i][temp:x])) # Add to Attila Y list
temp = x + 1
if (cnt == 3):
self.forceZ.append(float(
f1[i][temp:x])) # Add to Attila Z list
temp = x + 1
if (cnt == 4):
temp = x + 1
if (cnt == 5):
temp = x + 1
if (cnt == 6):
if (f1[17][29] == 'x'):
self.forceX2.append(float(
f1[i][temp:x])) # Add to Ryan X list
temp = x + 1
if (cnt == 7):
if (f1[17][34] == 'x'):
self.forceX2.append(float(
f1[i][temp:x])) # Add to Ryan X list
if (f1[17][32] == 'y'):
self.forceY2.append(float(
f1[i][temp:x])) # Add to Ryan Y list
temp = x + 1
if (cnt == 8):
if (f1[17][37] == 'y'):
self.forceY2.append(float(
f1[i][temp:x])) # Add to Ryan Y list
if (f1[17][35] == 'z'):
self.forceZ2.append(float(
f1[i][temp:x])) # Add to Ryan Z list
temp = x + 1
if (cnt == 9):
if (f1[17][40] == 'z'):
self.forceZ2.append(float(
f1[i][temp:x])) # Add to Ryan Z list
cnt = cnt + 1
#Get sampling rates to downsample data
video_SR = input("Input Video Sampling Rate (Hz): ")
data_SR = input("Input Data Sampling Rate (Hz): ")
print(" ")
#Set dwonsample trigger
downsample = int(data_SR) / int(video_SR)
up_samp = 0
if downsample - int(downsample) != 0:
up_samp = 1
#Declare lists for graphing data
self.graph = [0] * (self.total_frames)
self.graph = range(len(self.graph))
self.graphMax = [0] * (self.total_frames)
self.graphMin = [0] * (self.total_frames)
self.graphX = [0] * (self.total_frames)
self.graphY = [0] * (self.total_frames)
self.graphZ = [0] * (self.total_frames)
self.graphX2 = [0] * (self.total_frames)
self.graphY2 = [0] * (self.total_frames)
self.graphZ2 = [0] * (self.total_frames)
#Downsample data to video rate
if up_samp == 0:
cnt = 0
for i in range(len(self.time_abs)):
if cnt == downsample:
self.time_DS.append(self.time_abs[i])
self.forceX_DS.append(self.forceX[i])
self.forceY_DS.append(self.forceY[i])
self.forceZ_DS.append(self.forceZ[i])
self.forceX2_DS.append(self.forceX2[i])
self.forceY2_DS.append(self.forceY2[i])
self.forceZ2_DS.append(self.forceZ2[i])
cnt = 0
cnt += 1
#if resampling is needed
if up_samp == 1:
#find least common multiples and factors
lcm = np.lcm(int(data_SR), int(video_SR))
data_factor = int(lcm / int(data_SR))
video_factor = int(lcm / int(video_SR))
#resample data to a factor of data_factor
f_time = np.linspace(0, max(self.time_abs),
int(data_factor * len(self.time_abs)))
f_X = signal.resample(self.forceX, data_factor * len(self.forceX))
f_Y = signal.resample(self.forceY, data_factor * len(self.forceY))
f_Z = signal.resample(self.forceZ, data_factor * len(self.forceZ))
f_X2 = signal.resample(self.forceX2,
data_factor * len(self.forceX2))
f_Y2 = signal.resample(self.forceY2,
data_factor * len(self.forceY2))
f_Z2 = signal.resample(self.forceZ2,
data_factor * len(self.forceZ2))
#downsample data to video_factor
cnt = 0
for i in range(len(f_time)):
if cnt == video_factor:
self.time_DS.append(f_time[i])
self.forceX_DS.append(f_X[i])
self.forceY_DS.append(f_Y[i])
self.forceZ_DS.append(f_Z[i])
self.forceX2_DS.append(f_X2[i])
self.forceY2_DS.append(f_Y2[i])
self.forceZ2_DS.append(f_Z2[i])
cnt = 0
cnt += 1
#create data lists to grah
if self.contact_plate == "Attila":
self.list_creator(self.forceZ_DS, self.graphZ, self.frame_number1)
else:
self.list_creator(self.forceZ2_DS, self.graphZ2,
self.frame_number1)
count = 0
for x in range(self.frame_number1, self.total_frames - 1):
if count == len(self.time_DS):
break
self.graphX[self.frame_number1 +
count] = self.forceX_DS[self.frame_test + count]
self.graphY[self.frame_number1 +
count] = self.forceY_DS[self.frame_test + count]
self.graphX2[self.frame_number1 +
count] = self.forceX2_DS[self.frame_test + count]
self.graphY2[self.frame_number1 +
count] = self.forceY2_DS[self.frame_test + count]
if self.contact_plate == "Attila":
self.graphZ2[self.frame_number1 +
count] = self.forceZ2_DS[self.frame_test + count]
if self.contact_plate == "Ryan":
self.graphZ[self.frame_number1 +
count] = self.forceZ_DS[self.frame_test + count]
count += 1
#Reverse Y data if inversed
rev_Y = input("Reverse Y axis? (Y/N): ")
if rev_Y == "Y":
for x in range(len(self.graphY2)):
self.graphY[x] = -1 * self.graphY[x]
self.graphY2[x] = -1 * self.graphY2[x]
#Reverse X data if inversed
rev_X = input("Reverse X axis? (Y/N): ")
print(" ")
if rev_X == "Y":
for x in range(len(self.graphX2)):
self.graphX[x] = -1 * self.graphX[x]
self.graphX2[x] = -1 * self.graphX2[x]
#If trimming video, set data to trim points
if self.trim == "Y":
length = len(self.graph[self.trim_1:self.trim_2])
self.graph = self.time_DS[0:length]
self.graphX = self.graphX[self.trim_1:self.trim_2]
self.graphY = self.graphY[self.trim_1:self.trim_2]
self.graphZ = self.graphZ[self.trim_1:self.trim_2]
self.graphX2 = self.graphX2[self.trim_1:self.trim_2]
self.graphY2 = self.graphY2[self.trim_1:self.trim_2]
self.graphZ2 = self.graphZ2[self.trim_1:self.trim_2]
def resample(cls, frames, org_fs, dst_fs):
return signal.resample(frames, int((frames.size * dst_fs) / org_fs))
def resample_signal(msignal, n_sample):
return resample(msignal, n_sample)
def apply(self, data):
axis = data.ndim - 1
if data.shape[-1] > self.f:
return resample(data, self.f, axis=axis)
return data
globals()[VarName] = np.load(FilePath)
x_syn_peak, y_syn_peak = augment_train_set(xraw_test_peak[0:49, :],
y_test_peak[0:49, :], 100)
x_syn_per, y_syn_per = augment_train_set(xraw_test_per[0:49, :],
y_test_per[0:49, :], 100)
x_syn_osc, y_syn_osc = augment_train_set(xraw_test_osc[0:49, :],
y_test_osc[0:49, :], 100)
FolderPath_output = os.path.abspath(os.path.join(os.getcwd(),
"./output_data/"))
np.save(FolderPath_output + "/x_syn_peak.npy", x_syn_peak)
np.save(FolderPath_output + "/x_syn_per.npy", x_syn_per)
np.save(FolderPath_output + "/x_syn_osc.npy", x_syn_osc)
x_syn_peak = signal.resample(x_syn_peak, 10000, axis=1)
x_syn_peak = preprocessing.scale(x_syn_peak)
n_syn_norm = len(x_syn_peak)
x_syn_per = signal.resample(x_syn_per, 10000, axis=1)
x_syn_per = preprocessing.scale(x_syn_per)
x_syn_osc = signal.resample(x_syn_osc, 10000, axis=1)
x_syn_osc = preprocessing.scale(x_syn_osc)
#=============================================================================#
########### produce data: reality-augmented simulation faulty data ############
#=============================================================================#
n_syn = 2000 # synthetic time series to be produced
Syn_norm = ['peak', 'per']
FolderPath1 = ['v110', 'v90', 'v70', 'v50', 'v20']
FolderPath2 = ['normal', 'WF']
from numpy.random import normal
from numpy import log10, finfo, arange
import matplotlib.pyplot as plt
from scipy.fftpack import fft
from scipy.signal.windows import get_window
from scipy.signal import decimate, resample
eps = finfo("float").eps
def normalise(x):
return (x-min(x))/(max(x)-min(x))
sr = 44100
mean = 0
std = 1
N = 2**14
noise = normal(mean, std, size=N)
noisedec = resample(decimate(noise, 2), N)
window = get_window('blackmanharris', N, fftbins=False)
noisefft = abs(fft(noise * window))[:int(N/2)]
noisefftdec = abs(fft(noisedec * window))[:int(N/2)]
noisefft = 20*log10(normalise(noisefft) + eps)
noisefftdec = 20*log10(normalise(noisefftdec) + eps)
f = arange(int(N/2))/N*sr
fig, ax = plt.subplots(2, 1, figsize=(16,9))
ax[0].plot(f,noisefft)
ax[1].plot(f,noisefftdec)
plt.show()
def interpolate_peaks(data,
peaks,
sample_rate,
desired_sample_rate=1000.0,
working_data={}):
'''interpolate detected peak positions and surrounding data points
Function that enables high-precision mode by taking the estimated peak position,
then upsampling the peak position +/- 100ms to the specified sampling rate, subsequently
estimating the peak position with higher accuracy.
Parameters
----------
data : 1d list or array
list or array containing heart rate data
peaks : 1d list or array
list or array containing x-positions of peaks in signal
sample_rate : int or float
the sample rate of the signal (in Hz)
desired_sampled-rate : int or float
the sample rate to which to upsample.
Must be sample_rate < desired_sample_rate
Returns
-------
working_data : dict
working_data dictionary object containing all of heartpy's temp objects
Examples
--------
Given the output of a normal analysis and the first five peak-peak intervals:
>>> import heartpy as hp
>>> data, _ = hp.load_exampledata(0)
>>> wd, m = hp.process(data, 100.0)
>>> wd['peaklist'][0:5]
[63, 165, 264, 360, 460]
Now, the resolution is at max 10ms as that's the distance between data points.
We can use the high precision mode for example to approximate a more precise
position, for example if we had recorded at 1000Hz:
>>> wd = interpolate_peaks(data = data, peaks = wd['peaklist'],
... sample_rate = 100.0, desired_sample_rate = 1000.0, working_data = wd)
>>> wd['peaklist'][0:5]
[63.5, 165.4, 263.6, 360.4, 460.2]
As you can see the accuracy of peak positions has increased.
Note that you cannot magically upsample nothing into something. Be reasonable.
'''
assert desired_sample_rate > sample_rate, "desired sample rate is lower than actual sample rate \
this would result in downsampling which will hurt accuracy."
num_samples = int(0.1 * sample_rate)
ratio = sample_rate / desired_sample_rate
interpolation_slices = [(x - num_samples, x + num_samples) for x in peaks]
peaks = []
for i in interpolation_slices:
slice = data[i[0]:i[1]]
resampled = resample(
slice, int(len(slice) * (desired_sample_rate / sample_rate)))
peakpos = np.argmax(resampled)
peaks.append((i[0] + (peakpos * ratio)))
working_data['peaklist'] = peaks
return working_data
def scopeplot(x, width=800, height=400, range_=None, cmap=None, plot=None):
"""
Plot a signal using brightness to indicate density.
Parameters
----------
x : array_like, 1-D
The signal to be plotted
width, height : int, optional
The width and height of the output image in pixels. Default is
800×400.
range_ : float or 2-tuple of floats, optional
The vertical range of the plot. If a tuple, it is (xmin, xmax). If
a single number, the range is (-range, range). If None, it autoscales.
cmap : str or matplotlib.colors.LinearSegmentedColormap, optional
A matplotlib colormap for
Grayscale by default.
plot : bool or str or None, optional
If plot is None, the X image array is returned.
if plot is True, the image is plotted directly.
If plot is a string, it represents a filename to save the image to
using matplotlib's `imsave`.
Returns
-------
X : ndarray of shape (width, height)
A 2D array of amplitude 0 to 1, representing the density of the signal
at that point.
"""
if cmap is None:
cmap = 'gray'
x = asarray(x)
N = len(x)
# Add zeros to end to reduce circular Gibbs effects
MIN_PAD = 5 # TODO: what should this be? Seems subjective.
# Make input an optimal length for fast processing
pad_amount = next_fast_len(N + MIN_PAD) - N
x = pad(x, (0, pad_amount), 'constant')
# Resample such that signal evenly divides into chunks of equal length
new_size = int(round(_ceildiv(RS * N, width) * width / N * len(x)))
print('new size: {}'.format(new_size))
x = resample(x, new_size)
if not range_:
range_ = 1.1 * np.amax(np.abs(x))
if np.size(range_) == 1:
xmin, xmax = -range_, +range_
elif np.size(range_) == 2:
xmin, xmax = range_
else:
raise ValueError('range_ not understood')
spp = _ceildiv(N * RS, width) # samples per pixel
norm = 1 / spp
# Pad some zeros at beginning for overlap
x = pad(x, (spp // 2, 0), 'constant')
X = np.empty((width, height))
if spp % 2: # N is odd
chunksize = 2 * spp # (even)
else: # N is even
chunksize = 2 * spp + 1 # (odd)
print('spp: {}, chunk size: {}'.format(spp, chunksize))
for n in range(0, width):
chunk = x[n * spp:n * spp + chunksize]
assert len(chunk) # don't send empties
try:
h = _ihist(chunk, bins=height, range_=(xmin, xmax))
except ValueError:
print('argh', len(chunk))
else:
X[n] = h * norm
assert np.amax(X) <= 1.001, np.amax(X)
X = X**(0.4) # TODO: SUBJECTIVE
if isinstance(plot, str):
plt.imsave(plot, X.T, cmap=cmap, origin='lower', format='png')
elif plot:
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.imshow(X.T,
origin='lower',
aspect='auto',
cmap=cmap,
extent=(0, len(x), xmin, xmax),
interpolation='nearest')
# norm=LogNorm(vmin=0.01, vmax=300))
else:
return X
def stftMorph(self, x1, x2, fs, w1, N1, w2, N2, H1, smoothf, balancef):
"""
Morph of two sounds using the STFT
x1, x2: input sounds, fs: sampling rate
w1, w2: analysis windows, N1, N2: FFT sizes, H1: hop size
smoothf: smooth factor of sound 2, bigger than 0 to max of 1, where 1 is no smothing,
balancef: balance between the 2 sounds, from 0 to 1, where 0 is sound 1 and 1 is sound 2
returns y: output sound
"""
if (N2 / 2 * smoothf <
3): # raise exception if decimation factor too small
raise ValueError("Smooth factor too small")
if (smoothf > 1): # raise exception if decimation factor too big
raise ValueError("Smooth factor above 1")
if (balancef > 1
or balancef < 0): # raise exception if balancef outside 0-1
raise ValueError("Balance factor outside range")
if (H1 <= 0): # raise error if hop size 0 or negative
raise ValueError("Hop size (H1) smaller or equal to 0")
M1 = w1.size # size of analysis window
hM1_1 = math.floor(
(M1 + 1) // 2) # half analysis window size by rounding
hM1_2 = math.floor(M1 // 2) # half analysis window size by floor
L = x1.size // H1 # number of frames for x1
x1 = np.append(
np.zeros(hM1_2),
x1) # add zeros at beginning to center first window at sample 0
x1 = np.append(
x1, np.zeros(hM1_1)) # add zeros at the end to analyze last sample
pin1 = hM1_1 # initialize sound pointer in middle of analysis window
w1 = w1 / sum(w1) # normalize analysis window
M2 = w2.size # size of analysis window
hM2_1 = math.floor(
(M2 + 1) / 2) # half analysis window size by rounding
hM2_2 = math.floor(M2 / 2) # half analysis window size by floor2
H2 = x2.size // L # hop size for second sound
x2 = np.append(
np.zeros(hM2_2),
x2) # add zeros at beginning to center first window at sample 0
x2 = np.append(
x2, np.zeros(hM2_1)) # add zeros at the end to analyze last sample
pin2 = hM2_1 # initialize sound pointer in middle of analysis window
y = np.zeros(x1.size) # initialize output array
for l in range(L):
#-----analysis-----
mX1, pX1 = DFT.dftAnal(x1[pin1 - hM1_1:pin1 + hM1_2], w1,
N1) # compute dft
mX2, pX2 = DFT.dftAnal(x2[pin2 - hM2_1:pin2 + hM2_2], w2,
N2) # compute dft
#-----transformation-----
mX2smooth = resample(np.maximum(-200, mX2), int(
mX2.size * smoothf)) # smooth spectrum of second sound
mX2 = resample(mX2smooth,
mX1.size) # generate back the same size spectrum
mY = balancef * mX2 + (1 -
balancef) * mX1 # generate output spectrum
#-----synthesis-----
y[pin1 - hM1_1:pin1 + hM1_2] += H1 * DFT.dftSynth(
mY, pX1, M1) # overlap-add to generate output sound
pin1 += H1 # advance sound pointer
pin2 += H2 # advance sound pointer
y = np.delete(y, range(
hM1_2)) # delete half of first window which was added in stftAnal
y = np.delete(
y, range(y.size - hM1_1,
y.size)) # add zeros at the end to analyze last sample
# write morphed sound data
outputFullPath = outputPath + morphedFileName
#path4Writing = getRightPath(outputFullPath)
#sf.write(path4Writing, y, RATE)
# write sound file in env
commUtill.writeSoundFile(outputFullPath, y, RATE)
return outputFullPath
def main():
# Args
parser = argparse.ArgumentParser()
parser.add_argument("--model_name", default='CBN_10epch', help="Classifier model path")
parser.add_argument("--classifier", default='CBN', help="Choose classifier architecture, C, CBN")
args = parser.parse_args()
# Select training device
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# Load specified Classifier
if args.classifier == 'CBN':
net = ClassConvBN()
elif args.classifier == 'C':
net = ClassConv()
else:
net = ClassConv()
print('Bad Classifier option, running classifier C')
net.to(device)
# Load parameters from trained model
net.load_state_dict(torch.load('../../STEAD_CNN/models/' + args.model_name + '.pth'))
net.eval()
# Load Nevada data file 1
f = '../../Data_Utah/FORGE_78-32_iDASv3-P11_UTC190419001218.sgy'
# Read data
with segyio.open(f, ignore_geometry=True) as segy:
segy.mmap()
# Traces
data = segyio.tools.collect(segy.trace[:])
# Sampling frequency
fs = segy.header[0][117]
total = 0
tr_seismic, tr_noise = 0, 0
fil_seismic, fil_noise = 0, 0
seis_traces = []
seis_fil_traces = []
noise_traces = []
noise_fil_traces = []
# For every trace in the file
for idx, trace in enumerate(data):
# Filter
fil_trace = butter_bandpass_filter(trace, 0.5, 1, fs, order=5)
# Resample
resamp_trace = signal.resample(trace, 6000)
resamp_fil_trace = signal.resample(fil_trace, 6000)
# Normalize
resamp_trace = resamp_trace / np.max(np.abs(resamp_trace))
resamp_fil_trace = resamp_fil_trace / np.max(np.abs(resamp_fil_trace))
# Numpy to Torch
resamp_trace = torch.from_numpy(resamp_trace).to(device).unsqueeze(0)
resamp_fil_trace = torch.from_numpy(resamp_fil_trace).to(device).unsqueeze(0)
# Prediction
out_trace = net(resamp_trace.float())
out_fil_trace = net(resamp_fil_trace.float())
pred_trace = torch.round(out_trace.data).item()
pred_fil_trace = torch.round(out_fil_trace.data).item()
# Count traces
total += 1
if pred_trace:
tr_seismic += 1
seis_traces.append(idx)
else:
tr_noise += 1
noise_traces.append(idx)
if pred_fil_trace:
fil_seismic += 1
seis_fil_traces.append(idx)
else:
fil_noise += 1
noise_fil_traces.append(idx)
seis_tr_id = np.random.choice(seis_traces, 1)
seis_fil_tr_id = np.random.choice(seis_fil_traces, 1)
noise_tr_id = np.random.choice(seis_traces, 1)
noise_fil_tr_id = np.random.choice(seis_fil_traces, 1)
plt.figure()
plt.plot(data[seis_tr_id])
plt.savefig('seis_trace1.png')
plt.clf()
plt.plot(data[seis_fil_tr_id])
plt.savefig('seis_fil_trace1.png')
plt.clf()
plt.plot(data[noise_tr_id])
plt.savefig('noise_trace1.png')
plt.clf()
plt.plot(data[noise_fil_tr_id])
plt.savefig('noise_fil_trace1.png')
# Results
print(f'Inferencia Utah:\n\n'
f'File: FORGE_78-32_iDASv3-P11_UTC190419001218.sgy\n'
f'Total traces: {total}\n'
f'Predicted seismic: {tr_seismic}, predicted noise: {tr_noise}\n'
f'Predicted fil_seismic: {fil_seismic}, predicted fil_noise: {fil_noise}\n')
import scipy.signal as sg
import wave
from scipy import signal
# from master.scikits.talkbox.linpred.levinson_lpc import levinson,lpc
from scikits.talkbox.linpred.levinson_lpc import lpc
from scipy.io.wavfile import read
rate, signal = read("../Xe.wav")
signalA = []
for i in range(6000, 6512):
signalA.append(signal[i])
signalA = np.array(signalA)
p = 14
a, e, k = lpc(signalA, p, -1)
a = np.append(a, np.zeros(512-14))
data_freq = np.fft.fft(a, 512)
data_freq = sg.resample(data_freq,512)
magSpectrum = np.abs(data_freq)
# magDb = 1/magSpectrum
magDb = -np.log(magSpectrum)
signalArray = []
signalDK = []
for k in range(6000, 6900):
signalArray.append(signal[k])
signalArray = sg.resample(signalArray, 900)
for k in range(0, 149): # day la K 0..150
sig = 0
for m in range(0, 299):
sig = sig + np.abs(signalArray[m] - signalArray[ m - k]) # day la D(k)
signalDK.append(sig)
# plot.plot(signalDK)
plot.subplot(311)
def main():
cfg.device = torch.device('cuda')
torch.backends.cudnn.benchmark = False
max_per_image = 100
num_classes = 80 if cfg.dataset == 'coco' else 4
dictionary = np.load(cfg.dictionary_file)
colors = COCO_COLORS if cfg.dataset == 'coco' else DETRAC_COLORS
names = COCO_NAMES if cfg.dataset == 'coco' else DETRAC_NAMES
for j in range(len(names)):
col_ = [c * 255 for c in colors[j]]
colors[j] = tuple(col_)
print('Creating model and recover from checkpoint ...')
if 'hourglass' in cfg.arch:
model = exkp(n=5,
nstack=2,
dims=[256, 256, 384, 384, 384, 512],
modules=[2, 2, 2, 2, 2, 4],
num_classes=num_classes)
else:
raise NotImplementedError
model = load_demo_model(model, cfg.ckpt_dir)
model = model.to(cfg.device)
model.eval()
# Loading COCO validation images
annotation_file = '{}/annotations/instances_{}.json'.format(
cfg.data_dir, cfg.data_type)
coco = COCO(annotation_file)
# Load all annotations
cats = coco.loadCats(coco.getCatIds())
nms = [cat['name'] for cat in cats]
catIds = coco.getCatIds(catNms=nms)
imgIds = coco.getImgIds(catIds=catIds)
annIds = coco.getAnnIds(catIds=catIds)
all_anns = coco.loadAnns(ids=annIds)
for annotation in all_anns:
if annotation['iscrowd'] == 1 or type(
annotation['segmentation']) != list:
continue
img = coco.loadImgs(annotation['image_id'])[0]
image_path = '%s/images/%s/%s' % (cfg.data_dir, cfg.data_type,
img['file_name'])
w_img = int(img['width'])
h_img = int(img['height'])
if w_img < 1 or h_img < 1:
continue
polygons = annotation['segmentation'][0]
gt_bbox = annotation['bbox']
gt_x1, gt_y1, gt_w, gt_h = gt_bbox
contour = np.array(polygons).reshape((-1, 2))
# Downsample the contour to fix number of vertices
fixed_contour = resample(contour, num=cfg.num_vertices)
# Indexing from the left-most vertex, argmin x-axis
idx = np.argmin(fixed_contour[:, 0])
indexed_shape = np.concatenate(
(fixed_contour[idx:, :], fixed_contour[:idx, :]), axis=0)
clockwise_flag = check_clockwise_polygon(indexed_shape)
if not clockwise_flag:
fixed_contour = np.flip(indexed_shape, axis=0)
else:
fixed_contour = indexed_shape.copy()
fixed_contour[:, 0] = np.clip(fixed_contour[:, 0], gt_x1, gt_x1 + gt_w)
fixed_contour[:, 1] = np.clip(fixed_contour[:, 1], gt_y1, gt_y1 + gt_h)
contour_mean = np.mean(fixed_contour, axis=0)
contour_std = np.std(fixed_contour, axis=0)
# norm_shape = (fixed_contour - contour_mean) / np.sqrt(np.sum(contour_std ** 2.))
# plot gt mean and std
# image = cv2.imread(image_path)
# # cv2.ellipse(image, center=(int(contour_mean[0]), int(contour_mean[1])),
# # axes=(int(contour_std[0]), int(contour_std[1])),
# # angle=0, startAngle=0, endAngle=360, color=(0, 255, 0),
# # thickness=2)
# cv2.rectangle(image, pt1=(int(contour_mean[0] - contour_std[0] / 2.), int(contour_mean[1] - contour_std[1] / 2.)),
# pt2=(int(contour_mean[0] + contour_std[0] / 2.), int(contour_mean[1] + contour_std[1] / 2.)),
# color=(0, 255, 0), thickness=2)
# cv2.polylines(image, [fixed_contour.astype(np.int32)], True, (0, 0, 255))
# cv2.rectangle(image, pt1=(int(min(fixed_contour[:, 0])), int(min(fixed_contour[:, 1]))),
# pt2=(int(max(fixed_contour[:, 0])), int(max(fixed_contour[:, 1]))),
# color=(255, 0, 0), thickness=2)
# cv2.imshow('GT segments', image)
# if cv2.waitKey() & 0xFF == ord('q'):
# break
image = cv2.imread(image_path, cv2.IMREAD_COLOR)
original_image = image.copy()
height, width = image.shape[0:2]
padding = 127 if 'hourglass' in cfg.arch else 31
imgs = {}
for scale in cfg.test_scales:
new_height = int(height * scale)
new_width = int(width * scale)
if cfg.img_size > 0:
img_height, img_width = cfg.img_size, cfg.img_size
center = np.array([new_width / 2., new_height / 2.],
dtype=np.float32)
scaled_size = max(height, width) * 1.0
scaled_size = np.array([scaled_size, scaled_size],
dtype=np.float32)
else:
img_height = (new_height | padding) + 1
img_width = (new_width | padding) + 1
center = np.array([new_width // 2, new_height // 2],
dtype=np.float32)
scaled_size = np.array([img_width, img_height],
dtype=np.float32)
img = cv2.resize(image, (new_width, new_height))
trans_img = get_affine_transform(center, scaled_size, 0,
[img_width, img_height])
img = cv2.warpAffine(img, trans_img, (img_width, img_height))
img = img.astype(np.float32) / 255.
img -= np.array(
COCO_MEAN if cfg.dataset == 'coco' else DETRAC_MEAN,
dtype=np.float32)[None, None, :]
img /= np.array(COCO_STD if cfg.dataset == 'coco' else DETRAC_STD,
dtype=np.float32)[None, None, :]
img = img.transpose(
2, 0, 1)[None, :, :, :] # from [H, W, C] to [1, C, H, W]
# if cfg.test_flip:
# img = np.concatenate((img, img[:, :, :, ::-1].copy()), axis=0)
imgs[scale] = {
'image': torch.from_numpy(img).float(),
'center': np.array(center),
'scale': np.array(scaled_size),
'fmap_h': np.array(img_height // 4),
'fmap_w': np.array(img_width // 4)
}
with torch.no_grad():
segmentations = []
predicted_codes = []
start_time = time.time()
for scale in imgs:
imgs[scale]['image'] = imgs[scale]['image'].to(cfg.device)
output = model(imgs[scale]['image'])[-1]
segms, codes_ = ctsegm_scale_decode_debug(
*output,
torch.from_numpy(dictionary.astype(np.float32)).to(
cfg.device),
K=cfg.test_topk)
segms = segms.detach().cpu().numpy().reshape(
1, -1, segms.shape[2])[0]
codes_ = codes_.detach().cpu().numpy().reshape(
1, -1, codes_.shape[2])[0]
top_preds = {}
code_preds = {}
for j in range(cfg.num_vertices):
segms[:, 2 * j:2 * j + 2] = transform_preds(
segms[:, 2 * j:2 * j + 2], imgs[scale]['center'],
imgs[scale]['scale'],
(imgs[scale]['fmap_w'], imgs[scale]['fmap_h']))
segms[:, cfg.num_vertices * 2:cfg.num_vertices * 2 +
2] = transform_preds(
segms[:,
cfg.num_vertices * 2:cfg.num_vertices * 2 + 2],
imgs[scale]['center'], imgs[scale]['scale'],
(imgs[scale]['fmap_w'], imgs[scale]['fmap_h']))
segms[:, cfg.num_vertices * 2 + 2:cfg.num_vertices * 2 +
4] = transform_preds(
segms[:, cfg.num_vertices * 2 +
2:cfg.num_vertices * 2 + 4],
imgs[scale]['center'], imgs[scale]['scale'],
(imgs[scale]['fmap_w'], imgs[scale]['fmap_h']))
clses = segms[:, -1]
for j in range(num_classes):
inds = (clses == j)
top_preds[j + 1] = segms[inds, :cfg.num_vertices * 2 +
5].astype(np.float32)
top_preds[j + 1][:, :cfg.num_vertices * 2 + 4] /= scale
code_preds[j + 1] = codes_[inds, :]
segmentations.append(top_preds)
predicted_codes.append(code_preds)
segms_and_scores = {
j: np.concatenate([d[j] for d in segmentations], axis=0)
for j in range(1, num_classes + 1)
} # a Dict label: segments
codes_and_scores = {
j: np.concatenate([d[j] for d in predicted_codes], axis=0)
for j in range(1, num_classes + 1)
} # a Dict label: segments
scores = np.hstack([
segms_and_scores[j][:, cfg.num_vertices * 2 + 4]
for j in range(1, num_classes + 1)
])
if len(scores) > max_per_image:
kth = len(scores) - max_per_image
thresh = np.partition(scores, kth)[kth]
for j in range(1, num_classes + 1):
keep_inds = (segms_and_scores[j][:, cfg.num_vertices * 2 +
4] >= thresh)
segms_and_scores[j] = segms_and_scores[j][keep_inds]
codes_and_scores[j] = codes_and_scores[j][keep_inds]
# Use opencv functions to output a video
output_image = original_image
for lab in segms_and_scores:
for idx in range(len(segms_and_scores[lab])):
res = segms_and_scores[lab][idx]
c_ = codes_and_scores[lab][idx]
# for res in segms_and_scores[lab]:
contour, bbox, score = res[:-5], res[-5:-1], res[-1]
bbox[0] = np.clip(bbox[0], 0, w_img)
bbox[1] = np.clip(bbox[1], 0, h_img)
bbox[2] = np.clip(bbox[2], 0, w_img)
bbox[3] = np.clip(bbox[3], 0, h_img)
if score > cfg.detect_thres:
text = names[lab] # + ' %.2f' % score
# label_size = cv2.getTextSize(text, cv2.FONT_HERSHEY_COMPLEX, thickness=2, fontScale=0.5)
polygon = contour.reshape((-1, 2))
# use bb tools to draw predictions
color = random.choice(COLOR_WORLD)
bb.add(output_image, bbox[0], bbox[1], bbox[2],
bbox[3], text, color)
cv2.polylines(output_image, [polygon.astype(np.int32)],
True,
RGB_DICT[color],
thickness=2)
# color = (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))
# contour_mean = np.mean(polygon, axis=0)
# contour_std = np.std(polygon, axis=0)
# center_x, center_y = np.mean(polygon, axis=0).astype(np.int32)
# text_location = [bbox[0] + 1, bbox[1] + 1,
# bbox[1] + label_size[0][0] + 1,
# bbox[0] + label_size[0][1] + 1]
# cv2.rectangle(output_image, pt1=(int(bbox[0]), int(bbox[1])),
# pt2=(int(bbox[2]), int(bbox[3])),
# color=color, thickness=1)
# cv2.rectangle(output_image, pt1=(int(np.min(polygon[:, 0])), int(np.min(polygon[:, 1]))),
# pt2=(int(np.max(polygon[:, 0])), int(np.max(polygon[:, 1]))),
# color=(0, 255, 0), thickness=1)
# cv2.polylines(output_image, [polygon.astype(np.int32)], True, color, thickness=2)
# cv2.putText(output_image, text, org=(int(text_location[0]), int(text_location[3])),
# fontFace=cv2.FONT_HERSHEY_COMPLEX, thickness=2, fontScale=0.5,
# color=(255, 0, 0))
# cv2.putText(output_image, text, org=(int(bbox[0]), int(bbox[1])),
# fontFace=cv2.FONT_HERSHEY_COMPLEX, thickness=1, fontScale=0.5,
# color=color)
# show the histgram for predicted codes
fig = plt.figure()
plt.plot(np.arange(cfg.n_codes),
c_.reshape((-1, )),
color='green',
marker='o',
linestyle='dashed',
linewidth=2,
markersize=6)
plt.ylabel('Value of each coefficient')
plt.xlabel('All predicted {} coefficients'.format(
cfg.n_codes))
plt.title(
'Distribution of the predicted coefficients for {}'
.format(text))
plt.show()
cv2.imshow('Results', output_image)
if cv2.waitKey() & 0xFF == ord('q'):
break
