def apodise(image, border, order=8):
"""
Parameters
----------
image: np.ndarray
border: int, the size of the boreder in pixels
Note
----
The image is assumed to be of float datatype, no datatype management
is performed.
This is different from the original apodistation method,
which multiplied the image borders by a quater of a sine.
"""
# stackoverflow.com/questions/46211487/apodization-mask-for-fast-fourier-transforms-in-python
nx, ny = image.shape
# Define a general Gaussian in 2D as outer product of the function with itself
window = np.outer(
general_gaussian(nx, order, nx // 2 - border),
general_gaussian(ny, order, ny // 2 - border),
)
ap_image = window * image
return ap_image
def __init__(self, config, tool, **kwargs):
"""
Use a method for filtering the signal
Parameters
----------
config : traitlets.loader.Config
Configuration specified by config file or cmdline arguments.
Used to set traitlet values.
Set to None if no configuration to pass.
tool : ctapipe.core.Tool
Tool executable that is calling this component.
Passes the correct logger to the component.
Set to None if no Tool to pass.
kwargs
"""
super().__init__(config=config, parent=tool, **kwargs)
# Cleaning steps for plotting
self.stage_names = [
'0: raw',
'1: convolved (fast)',
'2: no pulse',
'3: smooth baseline',
'4: cleaned',
]
self.stages = {}
self.kernel_fast = general_gaussian(3, p=1.0, sig=1)
self.kernel_slow = general_gaussian(10, p=1.0, sig=32)
def __init__(self, config, tool, **kwargs):
"""
Use a method to smooth waveform
Parameters
----------
config : traitlets.loader.Config
Configuration specified by config file or cmdline arguments.
Used to set traitlet values.
Set to None if no configuration to pass.
tool : ctapipe.core.Tool
Tool executable that is calling this component.
Passes the correct logger to the component.
Set to None if no Tool to pass.
kwargs
"""
super().__init__(config=config, parent=tool, **kwargs)
if self.focus == 'pulse':
n_points = 20
pulsew = 1
self.kernel = general_gaussian(n_points, p=1.0, sig=pulsew)
elif self.focus == 'baseline':
n_points = 10
w = 32
self.kernel = general_gaussian(n_points, p=1.0, sig=w)
def filter_and_subtract(filtered, wavelength, window_size, sigma, pad=True):
mask = filtered.mask.copy()
orig = filtered.copy()
pad_size = 0
orig_begin = 0
orig_end = mask.size
if pad:
trimmed_filtered, orig_begin, orig_end = trim_array_from_mask(filtered, mask, buffer=0)
low_extent = np.ma.concatenate( (trimmed_filtered[:window_size/2+1], trimmed_filtered[:window_size/2+1][::-1]) )
high_extent = np.ma.concatenate( (trimmed_filtered[-window_size/2:][::-1], trimmed_filtered[-window_size/2:]) )
pad_size = low_extent.size
filtered = np.ma.concatenate( (low_extent, trimmed_filtered, high_extent) )
window = general_gaussian(window_size, p=2.5, sig=sigma, sym=True) + general_gaussian(window_size, p=0.5, sig=sigma*4, sym=True)
if np.any(filtered.mask) and np.any(~filtered.mask):
filtered[mask] = np.interp(wavelength[mask], wavelength[~mask], filtered[~mask])
filtered = fftconvolve(window, filtered)
filtered = (np.ma.average(orig) / np.ma.average(filtered)) * filtered
filtered = np.roll(filtered, -(window_size-1)/2)[:-(window_size-1)]
if pad:
new_filtered = np.ma.empty((mask.size, ), dtype=float)
new_filtered[:] = 0
new_filtered[orig_begin:orig_end+1] = filtered[pad_size:-pad_size]
new_filtered.mask = mask
filtered = new_filtered
return filtered
def getData(): #function called to get data
startTime = time.time()
raw650 = np.array(session[s1Vals])
raw950 = np.array(session[s2Vals])
# while True:
# if time.time() - startTime >= 5:
startTime = time.time()
print('got data')
working950 = reject_outliers(raw950)
working650 = reject_outliers(raw650)
sig950 = np.std(working950)
sig650 = np.std(working650)
print(sig650)
window950 = signal.general_gaussian(51, p=1.5, sig= sig950)
filtered950 = signal.fftconvolve(window950, working950)
filtered950 = (np.average(working950) / np.average(filtered950)) * filtered950
window650 = signal.general_gaussian(51, p=1.5, sig= sig650)
filtered650 = signal.fftconvolve(window650, working650)
filtered650 = (np.average(working650) / np.average(filtered650)) * filtered650
# filtered = np.roll(filtered, -25)
# plt.plot(working950)
# # plt.plot(window950)
# plt.plot(filtered950)
# plt.plot(raw650)
# plt.show()
print(filtered950)
print(working950)
concentration(filtered650,filtered950)
def get_reconstruction(tiff_path, discs, row, params):
# Read the tiff file
imgs = read_tiff(tiff_path)
if imgs[0].shape[0] != imgs[0].shape[1]:
print('The input image is not square. The image will be cropped to be a*a where `a` is the smallest dimension.')
a = min(imgs[0].shape[0], imgs[0].shape[1])
slice_x, slice_y = slice(0, a), slice(0, a)
imgs = [img[slice_x, slice_y] for img in imgs]
window = params['window']
a, p, sig = params['a'], params['p'], params['sig']
do_psd = params['do_psd']
# Apodization
width, height = imgs[0].shape # Images aren't supposed to have 3rd dimension
if window is not None and window.lower()=='gaussian':
w = np.outer(signal.general_gaussian(width, p=p, sig=sig), signal.general_gaussian(width, p=p, sig=sig))
elif window is not None and window.lower()=='tukey':
w = np.outer(signal.tukey(width, alpha=a), signal.tukey(height, alpha=a))
elif window is None or window.lower() is 'none':
w=1
imgs = [w*img for img in tqdm(imgs, desc='Processing Apodization', leave=False)]
# Periodic Smooth Decomposition
if do_psd:
imgs = [psd(img)[0] for img in tqdm(imgs, desc='Processing PSD', leave=False)]
imgs = [cp.array(img) for img in imgs] # Transfer to GPU
IMAGESIZE = imgs[0].shape[0]
scale = params['scale']
hres_size = (IMAGESIZE * scale, IMAGESIZE * scale)
# Remove keys not used by the reconstruction algo
prms = {k: params[k] for k in params.keys() - ['scale', 'do_psd', 'window', 'a', 'p', 'sig']}
# Reconstruction
print('Performing Reconstruction...', end='')
obj, pupil = reconstruct_v2(
imgs,
discs,
row,
hres_size,
**prms
)
print('Done!')
return obj, pupil, imgs
def __init__(self, config, tool, **kwargs):
"""
Use a method for filtering the signal
Parameters
----------
config : traitlets.loader.Config
Configuration specified by config file or cmdline arguments.
Used to set traitlet values.
Set to None if no configuration to pass.
tool : ctapipe.core.Tool
Tool executable that is calling this component.
Passes the correct logger to the component.
Set to None if no Tool to pass.
kwargs
"""
super().__init__(config=config, tool=tool, **kwargs)
# Cleaning steps for plotting
self.stages = {}
if self.t0:
self.log.info("User has set t0, extracted t0 will be overridden")
self.kernel = general_gaussian(10, p=1.0, sig=32)
def peakFinder(self, WLSTraces, windowParametersX, dataParametersX, medianFactorPF, stdFactor,
convWindow, convPower, convSig, minimaWindowPF):
countedPhotons = np.array([])
photonInd = []
print('Looking at '+str(len(WLSTraces.columns))+'traces.')
for i, thisTrace in enumerate(WLSTraces.columns):
data = np.array(-1*WLSTraces[thisTrace][windowParametersX['SIL']:windowParametersX['SIU']])
medCutoff = medianFactorPF*np.median(abs(data))
stdCutoff = stdFactor*np.std(data[np.where(abs(data) < medCutoff)])
window = signal.general_gaussian(convWindow, p=convPower, sig=convSig)
filtered = signal.fftconvolve(window, data)
filtered = (np.average(data) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -1*int((convWindow-1)/2))
peakind = signal.argrelmax(filtered, order=minimaWindowPF)[0]
newind = peakind[np.where(filtered[peakind] > stdCutoff)]
photonInd.append(newind)
countedPhotons = np.append(countedPhotons, len(newind))
if ((i % 100) == 0):
print('Working on traces '+self.c.blue(str(i)+' - '+str(i+99))+' for peak detection...')
print('Total photons counted: '+self.c.lightgreen(str(sum(countedPhotons))))
print('Average number of photons per trace: '+self.c.lightgreen(str(sum(countedPhotons)/len(WLSTraces.columns))))
print('Median number of photons: '+self.c.lightgreen(str(np.median(countedPhotons))))
return photonInd, countedPhotons
def detect(self, sampFreq, snd):
if type(snd[0]) != np.int16:
sndBackup = [snd[i][0] for i in range(len(snd))]
snd = np.array(sndBackup)
width = np.array(
range(int(0.15 * sampFreq), int(3 * sampFreq), int(
0.05 * sampFreq))) / 100
snd = snd / 1000.0
power = abs(snd)
power = self.myfilter(power, sampFreq)
powerSample = self.downSampling(power, 100)
window = signal.general_gaussian(500, p=0.5, sig=500)
filtered = signal.fftconvolve(window, powerSample)
filtered = (np.average(powerSample) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -25)
peaks = signal.find_peaks_cwt(filtered, width)
peakTimeDiff = [peaks[i] - peaks[i - 1] for i in range(1, len(peaks))]
return (np.std(peakTimeDiff) / np.mean(peakTimeDiff)) < 0.4
def Slownoise_Smooth(self,data):
"""
generates smoothed signal from peak-less signal and raw_data
takes: peak-less signal , raw_data
returns: smoothed signal 1
"""
slownoise = self.RemovePeaksBasedOnRMS(data,3)
nsamples = 500
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples/5)
slownoise_smooth = sig.fftconvolve(slownoise, window, "same")
slownoise_smooth = (np.average(slownoise) / np.average(slownoise_smooth)) * slownoise_smooth
self.filtered_signal_1 = data - slownoise_smooth
self.peakless = slownoise
'''
plt.figure()
plt.title('slownoise')
plt.xlabel('Time [s]')
plt.ylabel('Voltage [V]')
plt.plot(self.time,data,'grey')
plt.plot(self.time,slownoise,'red')
plt.title('slownoise_smooth')
plt.plot(self.time,slownoise_smooth,'blue')
#plt.figure()
plt.xlabel('Time [s]')
plt.ylabel('Voltage [V]')
#plt.title('raw data ,filtered signal 1')
#plt.plot(self.time,data,'grey')
#plt.plot(self.time,self.filtered_signal_1)
plt.show()
'''
return self.filtered_signal_1
def plot_cm(self, minVal, maxVal, y_axmin=0, y_axmax=255, array=None, smooth=False, peaks=False, mind=1, ind=1, canny=False):
#Create Plot of Contour Mean Intensities.
if self.all_cm == None or self.filtered_contours == None:
self.big_cm(canny=canny, minVal=minVal, maxVal=maxVal, array=array)
c_m = self.all_cm
if peaks and not smooth:
data = c_m[:,ind,1]
window = signal.general_gaussian(4, p=0.5, sig=100)
filtered = signal.fftconvolve(window, data)
filtered = (np.average(data)/np.average(filtered))*filtered
detect_peaks(filtered, show=True, mpd = mind, y_axmin=y_axmin, y_axmax=y_axmax)
#Source: http://nbviewer.ipython.org/github/demotu/BMC/blob/master/notebooks/DetectPeaks.ipynb
if smooth:
xnew=np.linspace(0,len(c_m[:,:,1]),len(c_m[:,:,1]))
y1=c_m[:,ind,1]#for now must be specified
f=interp1d(xnew,y1,kind='cubic')(xnew)
plt.plot(f)
plt.ylabel('Mean Intensity')
plt.show()
if not smooth and not peaks:
plt.plot(c_m[:,:,1])
plt.ylabel('Mean Intensity')
plt.show()
def Slownoise_Smooth(self,data,ite):
"""
generates smoothed signal from peak-less signal and raw_data
takes: peak-less signal , raw_data
returns: smoothed signal 1
"""
slownoise,level = self.RemovePeaksBasedOnRMS3(data)
meanzero = np.mean(slownoise)
self.peakless = slownoise - meanzero #peakless signal - pedestal mean
data_at_0 = data - meanzero #subtract mean of pedestal from data
# not given that trace is always with same offset // gaincalc only takes into account meanzero of last segment
nsamples = 700 #determines width of smoothing gaussian (/5) = sigma of gauss
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples/5)
slownoise_smooth = sig.fftconvolve(self.peakless, window, "same")
slownoise_smooth = (np.average(self.peakless) / np.average(slownoise_smooth)) * slownoise_smooth
self.filtered_signal_1 = data_at_0 - slownoise_smooth
self.ReportProgress(ite,int(float(self.calcsegmentspro)/100*self.ranger))
#print 'datamean ',np.mean(data)
#print 'meanzero ',meanzero
#print 'data0 ',np.mean(data_at_0)
return self.filtered_signal_1
def audio_resample_tf(y,
orig_sr,
target_sr,
res_type='kaiser_best',
fix=True,
scale=False,
use_smoothing=True,
use_bicubic=False):
ratio = float(target_sr) / orig_sr
y_length = y.get_shape().as_list()[1]
n_samples = int(np.ceil(y_length * ratio))
if use_smoothing:
window = general_gaussian(5, 0.5, True).astype(np.float32)
window /= window.sum()
window_tf = tf.constant(window[:, np.newaxis, np.newaxis, np.newaxis],
dtype=tf.float32)
y = tf.nn.conv2d(y, window_tf, strides=[1, 1, 1, 1], padding="SAME")
if (not use_bicubic) and (np.mod(np.log2(float(orig_sr) / target_sr), 1)
== 0.0):
for i in range(int(np.log2(float(orig_sr) / target_sr)) - 1):
y = y[:, ::2, :, :]
y = tf.nn.conv2d(y,
window_tf,
strides=[1, 1, 1, 1],
padding="SAME")
y_hat = y[:, ::2, :, :]
else:
y_hat = bicubic_interp_2d(y, [n_samples, 1], endpoint=False)
if scale:
y_hat /= np.sqrt(ratio)
return y_hat
def Apod(im, size, power, sigma):
av = im.mean()
im = im - av
gen_gaussian = general_gaussian(size, power, sigma)
window = np.outer(gen_gaussian, gen_gaussian)
apod = im * window
apod = apod + av
return apod
def Slownoise_Smooth(self, data, ite):
"""
generates smoothed signal from peak-less signal and raw_data
takes: peak-less signal , raw_data
returns: smoothed signal 1
"""
slownoise, level = self.RemovePeaksBasedOnRMS3(data)
meanzero = np.mean(slownoise)
self.peakless = slownoise - meanzero #peakless signal - pedestal mean
meanpeakless = np.mean(self.peakless)
#print meanpeakless
#print meanzero, ' ',meanpeakless
data_at_0 = data - meanzero #subtract mean of pedestal from data
# not given that trace is always with same offset // gain-calc only takes into account meanzero of last segment
nsamples = 700 #determines width of smoothing gaussian (/5) = sigma of gauss
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples / 5)
slownoise_smooth = sig.fftconvolve(self.peakless, window, "same")
slownoise_smooth = (np.average(self.peakless) /
np.average(slownoise_smooth)) * slownoise_smooth
self.filtered_signal_1 = data_at_0 - slownoise_smooth
if ite == 0:
fig3, (ax1, ax2) = plt.subplots(1, 2, sharey=False)
rawdat = self.raw_data * 1e3 #mV
ax1.set_title('Slow Noise')
ax1.set_xlabel('Time [$\mu$s]')
ax1.set_xlim(0, 100)
ax1.set_ylabel('Voltage [mV]')
ax1.plot(self.time, data, 'b', label='Raw Data')
ax1.plot(self.time,
slownoise_smooth,
'w',
label='Slownoise Smooth')
ax2.set_title('Zoom')
ax2.set_xlabel('Time [$\mu$s]')
ax2.set_xlim(45, 46)
ax2.set_ylim(-0.01, 0.04)
ax2.set_ylabel('Voltage [mV]')
ax2.plot(self.time, data, 'b', label='Raw Data')
ax2.plot(self.time,
slownoise_smooth,
'k',
label='Slownoise Smooth')
ax2.legend()
fig3.savefig(self.data_file_path + 'SlowNoiseSmoothZoom.pdf',
format='pdf',
bbox_inches='tight')
#plt.show()
self.ReportProgress(ite, self.ranger)
return self.filtered_signal_1
def __init__(self, config, tool, **kwargs):
super().__init__(config=config, tool=tool, **kwargs)
# Cleaning steps for plotting
self.stages = {}
self.kernel = general_gaussian(10, p=1.0, sig=32)
self.extractor = self.get_extractor()
def __init__(self, config, tool, **kwargs):
super().__init__(config=config, tool=tool, **kwargs)
# Cleaning steps for plotting
self.stages = {}
self.kernel = general_gaussian(10, p=1.0, sig=32)
self.extractor = self.get_extractor()
def findbiggestGauss(x, y, fitwindow=20):
window = signal.general_gaussian(11, p=0.5, sig=10)
filtered = signal.fftconvolve(window, y)
filtered = (np.average(y) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -5)[:len(y)]
window = signal.general_gaussian(11, p=0.5, sig=200)
deriv = np.gradient(filtered)
dfiltered = signal.fftconvolve(window, deriv)
dfiltered = (np.average(deriv) / np.average(dfiltered)) * dfiltered
dfiltered = np.roll(dfiltered, -15)[:len(deriv)]
zeros = np.sign(dfiltered)
zeros[zeros == 0] = -1 # replace zeros with -1
zeros = np.where(np.diff(zeros))[0]
max = np.argmax(filtered[zeros])
max = zeros[max]
# print(x[max])
initial_guess = (filtered[max], x[max], 10)
minind = max - fitwindow
if minind < 0:
minind = 0
maxind = max + fitwindow
if maxind >= len(x):
maxind = len(x) - 1
indices = np.linspace(minind, maxind, fitwindow * 2 + 1, dtype=int)
# popt, pcov = curve_fit(gauss, x[indices], y[indices], p0=initial_guess)
# perr = np.sqrt(np.diag(pcov))
def err_fun(p):
fit = gauss(x[indices], *p)
diff = np.abs(y[indices] - fit)**2
return np.sum(diff)
minimizer_kwargs = {"method": "SLSQP", "tol": 1e-12}
res = basinhopping(err_fun,
initial_guess,
minimizer_kwargs=minimizer_kwargs,
niter=50)
# res = minimize(err_fun, start, method='SLSQP',tol=1e-12,bounds = bnds, options={ 'disp': True, 'maxiter': 500})
# res = minimize(err_fun, initial_guess, method='L-BFGS-B', jac=False, options={'disp': False, 'maxiter': 500})
# res = minimize(err_fun, start, method='nelder-mead', options={'xtol': 1e-2, 'disp': True})
popt = res.x
fitted = gauss(x, popt[0], popt[1], popt[2])
return popt, fitted
def ppeaks(self, minVal, maxVal, ind=1, mph=1, mpd=1, mph2=-95, mpd2=1, thresh=0,
array=None, fps=40, calc_widths=False, minWidth=4, maxWidth=50, avg_fluor=True,
show=False, show_vl=False, deltax=25, y_axmin=0, y_axmax=2.2, stddev=None,lam=100, p=.001, niter=6, normed=False):
''' Peak Detection of Contour Mean Intensities.
http://nbviewer.ipython.org/github/demotu/BMC/blob/master/notebooks/DetectPeaks.ipynb
mph : detect peaks that are greater than minimum peak height.
mph2: same, just for valleys (and negative!)
mpd : detect peaks that are at least separated by minimum peak distance.
mpd2 : same, but valleys
threshold : detect peaks (valleys) that are greater (smaller) than `threshold`
in relation to their immediate neighbors.
Seems to work: (see iPython for other population params.)
for i in range(length):
q.ppeaks(ind=i, mph=85, mpd=15,thresh=0.15, minVal=70, maxVal=175)
Normed divides each bit of data by its baseline.
'''
if self.all_cm == None or self.filtered_contours == None:
self.big_cm(canny=True, array=array, minVal=minVal, maxVal=maxVal)
if stddev==None:
c_m = self.all_cm
data = c_m[:,ind,1]
if stddev!=None: #this means look at nonblinky cells (stddev will be low, coded earlier.)
if self.stdd==None:
self.create_std(stddev=stddev, avg_fluor=avg_fluor)
ar = self.stdd
data = ar[ind,:]
window = signal.general_gaussian(4, p=0.5, sig=100)
filtered = signal.fftconvolve(window, data)
filtered = (np.average(data)/np.average(filtered))*filtered
filtered = filtered[4:-5] #truncates bad boundaries from fftconvolve
original_filtered = np.copy(filtered)
bline=self.baseline_als(filtered, lam=lam, p=p, niter=niter)
if normed:
filtered=filtered/bline
peaks = detect_peaks(x=filtered, mpd=mpd, mph=mph, threshold=thresh, edge='rising',
show=True, y_axmin=0, y_axmax=y_axmax)
#collect the normed traces of all cells.
self.normed_trace.append(filtered)
y_peaks = []
for ind in peaks:
y_peaks.append(filtered[ind])
if calc_widths==True:
self.p_width(peaks=peaks, bline=bline, filtered=filtered, minWidth=minWidth, maxWidth=maxWidth, cell_ind=ind, show=show, data=data, deltax=deltax, normed=normed)
frames = np.arange(0,len(filtered))
if show_vl==True:
plt.plot(frames, original_filtered, 'm', frames, bline,'b--') #filtered-->original_filtered
plt.show()
def peak_finder(times, residuals, errors, params, n_peaks=4, plots=False, verbose=False):
# http://stackoverflow.com/a/25666951
# Convolve residuals with a gaussian, find relative maxima
n_points_kernel = 100
window = signal.general_gaussian(n_points_kernel+1, p=1, sig=12)
filtered = signal.fftconvolve(window, residuals)
filtered = (np.average(residuals) / np.average(filtered)) * filtered
filtered = np.roll(filtered, int(-n_points_kernel/2))
maxes = signal.argrelmax(filtered)[0]
maxes = maxes[maxes < len(residuals)]
maxes = maxes[residuals[maxes] > 0]
lower_t_bound, upper_t_bound = get_in_transit_bounds(times, params)
maxes_in_transit = maxes[(times[maxes] < upper_t_bound) &
(times[maxes] > lower_t_bound)]
if len(maxes_in_transit) == 0:
if verbose:
print('no maxes found')
return None
peak_times = times[maxes_in_transit]
peak_amplitudes = residuals[maxes_in_transit]#len(peak_times)*[3*np.std(residuals)]
peak_sigmas = len(peak_times)*[4./60/24] # 5 min
input_parameters = np.vstack([peak_amplitudes, peak_times, peak_sigmas]).T.ravel()
# result = optimize.fmin(chi2, input_parameters, args=(times, residuals, errors))
result = optimize.fmin_bfgs(chi2, input_parameters, disp=False,
args=(times, residuals, errors))
#print(result, result == input_parameters)
if plots:
fig, ax = plt.subplots(3, 1, figsize=(8, 10), sharex=True)
ax[0].errorbar(times, residuals, fmt='.', color='k')
[ax[0].axvline(t) for t in times[maxes_in_transit]]
ax[0].plot(times, summed_gaussians(times, input_parameters), 'r')
#ax[1].errorbar(times, gaussian_model, fmt='.', color='r')
ax[0].axhline(0, color='k', ls='--')
ax[0].set_ylabel('Transit Residuals')
ax[1].errorbar(times, residuals, fmt='.', color='k')
ax[1].plot(times, summed_gaussians(times, result), 'r')
#ax[1].errorbar(times, gaussian_model, fmt='.', color='r')
ax[1].axhline(0, color='k', ls='--')
ax[1].set_ylabel('Residuals')
ax[2].errorbar(times, residuals - summed_gaussians(times, result), fmt='.', color='k')
#ax[1].errorbar(times, gaussian_model, fmt='.', color='r')
ax[2].axhline(0, color='k', ls='--')
ax[2].set_ylabel('Residuals')
fig.tight_layout()
plt.show()
return result
def fftfilter(y):
window = signal.general_gaussian(11, p=0.5, sig=10)
filtered = signal.fftconvolve(window, y)
filtered = (np.average(y) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -5)[:len(y)]
# filtered = filtered[:1024]
#peakidx = signal.find_peaks_cwt(filtered, np.arange(1, 10), noise_perc=0.01)
#plt.plot(x, y)
#plt.plot(x, filtered, "g")
#plt.plot(x[peakidx], y[peakidx], "rx")
#plt.show()
return filtered
def noise_pnorm(R, p, n, train_data):
train_n = np.zeros((200, 204))
for i in enumerate(train_data):
epsilon = signal.general_gaussian(n, p, sig=p)
epsilon[abs(epsilon) < 1e-5] = 0
sign = [(-1)**np.random.randint(2) for i in range(n)]
x = np.multiply(epsilon, sign)
z = (np.random.uniform(0, 1))**(1 / n)
x_pnorm = np.linalg.norm(x, ord=p)
y = R * z * x / x_pnorm
train_n[i[0]] = train_data[i[0]] + y
return train_n
def __init__(self, config=None, tool=None, **kwargs):
super().__init__(config=config, tool=tool, **kwargs)
# Cleaning steps for plotting
self.stages = {}
self.stage_names = [
'0: raw', '1: baseline_sub', '2: no_pulse', '3: smooth_baseline',
'4: smooth_wf', '5: cleaned'
]
self.kernel = general_gaussian(10, p=1.0, sig=32)
self.extractor = self.get_extractor()
def work(self, input_items, output_items):
in0 = input_items[0]
out0 = output_items[0]
window = signal.general_gaussian(self.m, p=self.p, sig=self.sigma)
for i in range(0, len(in0)):
filtered = signal.fftconvolve(window, in0[i])
filtered = (numpy.average(in0[i]) /
numpy.average(filtered)) * filtered
out0[i][:] = numpy.roll(filtered, -self.m / 2)[:len(in0[i])]
return len(in0)
def Fastnoise_Smooth(self,filtered_signal_1):
"""
smoothes noise-less signal again
takes: noise-less signal
returns: smoothed noise-less signal
peak detection more reliable
"""
nsamples = 10#20
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples/5)
self.filtered_signal_2 = sig.fftconvolve(filtered_signal_1, window, "same")
self.filtered_signal_2 = (np.average(filtered_signal_1) / np.average(self.filtered_signal_2)) * self.filtered_signal_2
return self.filtered_signal_2
def raw_ctd_filter(input_array=None,
filter_type='triangle',
win_size=24,
parameters=None):
"""raw_ctd_filter function
Function takes NUMPY array
of raw ctd data and returns filtered data. This function also needs
one of three filter types (boxcar, gaussian, triangle) as well as
window size.
Args:
param1 (ndarray): Numpy ndarray with predefined header with at
param2 (str): One of three tested filter types
boxcar, gaussian_std, triangle.
default is triangle
param3 (int): A window size for the filter. Default is 24, which
is the number of frames per second from a SBE9+/11 CTD/Dech unit.
param4 (ndarray): parameters the dtype names used in filtering the
analytical inputs.
Returns:
Narray: The return value is a matrix of filtered ctd data with
the above listed header values.
"""
if input_array is None:
print("In raw_ctd_filter: No data array.")
return
else:
return_array = input_array
if parameters is None:
print("In raw_ctd_filter: Empty parameter list.")
else:
for p in parameters:
if filter_type is 'boxcar':
win = sig.boxcar(win_size)
return_array[str(p)] = sig.convolve(
input_array[str(p)], win, mode='same') / len(win)
elif filter_type is 'gaussian':
sigma = np.std(arr)
win = sig.general_gaussian(win_size, 1.0, sigma)
return_array[str(p)] = sig.convolve(
input_array[str(p)], win, mode='same') / (len(win))
elif filter_type is 'triangle':
win = sig.triang(win_size)
return_array[p] = 2 * sig.convolve(
input_array[p], win, mode='same') / len(win)
return return_array
def general_gaussian_wf_vs_nf(μ, σ, p):
"""
Plot a general Gaussian distribution of weight factor vs number factor.
:param μ: mean
:param σ: standard deviation
:param p: shape
:return: fig, ax
"""
number_of_points = 100
numbers = general_gaussian(number_of_points, p,
number_of_points * σ / (2 * μ))
weights = np.linspace(0, 2 * μ, number_of_points)
return graph_wf_vs_nf(numbers, weights)
def __init__(self, config, tool, **kwargs):
super().__init__(config=config, tool=tool, **kwargs)
# Cleaning steps for plotting
self.stages = {}
self.stage_names = ['0: raw',
'1: baseline_sub',
'2: no_pulse',
'3: smooth_baseline',
'4: smooth_wf',
'5: cleaned']
self.kernel = general_gaussian(10, p=1.0, sig=32)
self.extractor = self.get_extractor()
def Slownoise_Smooth(self, data, ite):
"""
generates smoothed signal from peak-less signal and raw_data
takes: peak-less signal , raw_data
returns: smoothed signal 1
"""
rmsn = 2
slownoise, level = self.RemovePeaksBasedOnRMS3(data, rmsn)
nsamples = 700 #determines width of smoothing gaussian (/5) = sigma of gauss
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples / 5)
slownoise_smooth = sig.fftconvolve(slownoise, window, "same")
#slownoise_smooth = (np.average(slownoise) / np.average(slownoise_smooth)) * slownoise_smooth
slownoise_smooth = (np.average(slownoise) /
np.average(slownoise_smooth)) * slownoise_smooth
self.filtered_signal_1 = data - slownoise_smooth
self.peakless = slownoise
if ite == 1:
fig8 = plt.figure(8)
ax8 = fig8.add_axes([0.1, 0.1, 0.8, 0.8])
ax8.set_title('Filter Levels')
ax8.set_xlabel('Time [us]')
ax8.set_ylabel('Voltage [V]')
#ax8.set_xlim(0,100)
rawdat = ax8.plot(self.time, data, 'grey', label='Raw Data')
slown = ax8.plot(self.time, slownoise, 'red', label='slow Noise')
plt.plot((np.amin(self.time), np.amax(self.time)),
(level * rmsn, level * rmsn),
'b--',
label='Cut level')
ax8.set_title('Smoothed slow Noise')
slowns = ax8.plot(self.time,
slownoise_smooth,
'blue',
label='Smoothed slow Noise')
ax8.legend()
#plt.show()
fig8.savefig(self.data_file_path + 'FilterLevels.ps',
format='ps',
bbox_inches='tight')
#ax8.legend_.remove()
#fig8.clf()
self.ReportProgress(ite)
return self.filtered_signal_1
def smooth_with_gaussian(time_series, sfreq=1000, gaussian_width=50, N=1000):
# gaussian_width in samples
# ---------------------
import math
from scipy import signal
norm_factor = np.sqrt(
2 * math.pi * gaussian_width**
2) / sfreq # sanity check: norm_factor = gaussian_window.sum()
gaussian_window = signal.general_gaussian(
M=N, p=1, sig=gaussian_width) # generate gaussian filter
norm_factor = (gaussian_window / sfreq).sum()
smoothed_time_series = np.convolve(time_series,
gaussian_window / norm_factor,
mode="full") # smooth
smoothed_time_series = smoothed_time_series[int(round(
N / 2)):-(int(round(N / 2)) - 1)] # trim ends
return smoothed_time_series
def _find_grid_pins(resampled_row, fiber, exposure):
data = resampled_row['flux']+resampled_row['sky']
wavelength = resampled_row['wavelength']
window = general_gaussian(21, p=0.5, sig=3)
filtered = data[:peak_seek_end].copy()
if np.any(filtered.mask) and np.any(~filtered.mask):
filtered[filtered.mask] = np.interp(wavelength[filtered.mask], wavelength[~filtered.mask], filtered[~filtered.mask])
filtered = fftconvolve(window, filtered)
filtered = (np.ma.average(data[:peak_seek_end]) / np.ma.average(filtered)) * filtered
filtered = np.roll(filtered, -10)[:peak_seek_end]
filtered_peak_inds = np.array(argrelextrema(filtered, np.ma.greater))
filtered_peak_wlens = (filtered_peak_inds*0.975)+3500.26
peak_set = [fiber, exposure]
for target in pin_peaks:
peak_set.append(_find_closest_peak(target, wavelength, data))
return tuple(peak_set)
def smoothing_to_get_peaks(sgnl, x_sgnl, param):
if param['do']:
window = signal.general_gaussian(param['M'],
p=param['p'],
sig=param['sig'])
ftrd_sgnl = signal.fftconvolve(window, sgnl)
ftrd_sgnl = (np.average(sgnl) / np.average(ftrd_sgnl)) * ftrd_sgnl
ftrd_sgnl = ftrd_sgnl[:((-1) * param['M']) + 1]
else:
ftrd_sgnl = sgnl
if param['plot']:
plt.plot(x_sgnl, ftrd_sgnl)
plt.xlabel('Frequency [Hz]', fontsize=10)
plt.ylabel('Amplitude', fontsize=10)
plt.title("Gausian filtering of Freq. Domain FFT", fontsize=12)
plt.show()
return ftrd_sgnl
def Fastnoise_Smooth(self,filtered_signal_1):
"""
smoothes noise-less signal again
takes: noise-less signal
returns: smoothed noise-less signal
peak detection more reliable
"""
sig_smooth_ns = 4 # Depends on pulse shape...
win_smooth_ns = sig_smooth_ns * 5 #... possibly FWHM * 1.25
samples_per_ns = 2.5
window = sig.general_gaussian(win_smooth_ns*samples_per_ns
, p=1.0, sig=sig_smooth_ns*samples_per_ns)
self.filtered_signal_2 = sig.fftconvolve(filtered_signal_1, window, "same")
self.filtered_signal_2 = (np.average(filtered_signal_1) / np.average(self.filtered_signal_2)) * self.filtered_signal_2
return self.filtered_signal_2
def get_onset_envelope(Sxx, ts, timestep=None, draw=False):
"""
Gets the strength of note onsets throughout a recording given a
spectrogram. Calculates spectral flux and then uses a Gaussian
filter to find the onset strength envelope.
"""
# calculate spectral flux
dt = ts[1] - ts[0]
if timestep is None:
timestep = 0
shift_len = int(timestep / dt) + 1
n, m = Sxx.shape
blank = np.zeros((n, shift_len))
old = np.hstack((blank, Sxx))
new = np.hstack((Sxx, blank))
diff = (new - old)[:, :m]
diff = onset_filter(diff)
diff = np.sum(diff, axis=0)
diff[:shift_len] = 0
if draw:
plt.plot(diff)
# gaussian filter
diff += 1
size = 2
window = signal.general_gaussian(size * 2 + 1, p=1, sig=20)
filtered = signal.fftconvolve(window, diff)
filtered = (np.average(diff) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -size)
diff = filtered
diff = diff[:m]
l = np.average(diff) - np.std(diff)
smallest = (np.max(diff) - np.min(diff)) * 0.1 + np.min(diff)
l = smallest
diff[diff < l] = l
if draw:
plt.plot(diff)
plt.show()
return diff
def _find_grid_pins(resampled_row, fiber, exposure):
data = resampled_row['flux'] + resampled_row['sky']
wavelength = resampled_row['wavelength']
window = general_gaussian(21, p=0.5, sig=3)
filtered = data[:peak_seek_end].copy()
if np.any(filtered.mask) and np.any(~filtered.mask):
filtered[filtered.mask] = np.interp(wavelength[filtered.mask],
wavelength[~filtered.mask],
filtered[~filtered.mask])
filtered = fftconvolve(window, filtered)
filtered = (np.ma.average(data[:peak_seek_end]) /
np.ma.average(filtered)) * filtered
filtered = np.roll(filtered, -10)[:peak_seek_end]
filtered_peak_inds = np.array(argrelextrema(filtered, np.ma.greater))
filtered_peak_wlens = (filtered_peak_inds * 0.975) + 3500.26
peak_set = [fiber, exposure]
for target in pin_peaks:
peak_set.append(_find_closest_peak(target, wavelength, data))
return tuple(peak_set)
def Slownoise_Smooth(self, data, ite):
"""
generates smoothed signal from peak-less signal and raw_data
takes: peak-less signal , raw_data
returns: smoothed signal 1
"""
slownoise, level = self.RemovePeaksBasedOnRMS3(data)
meanzero = np.mean(slownoise)
self.peakless = slownoise - meanzero #peakless signal - pedestal mean
meanpeakless = np.mean(self.peakless)
#print meanpeakless
#print meanzero, ' ',meanpeakless
data_at_0 = data - meanzero #subtract mean of pedestal from data
# not given that trace is always with same offset // gain-calc only takes into account meanzero of last segment
nsamples = 700 #determines width of smoothing gaussian (/5) = sigma of gauss
window = sig.general_gaussian(nsamples, p=1.0, sig=nsamples / 5)
slownoise_smooth = sig.fftconvolve(self.peakless, window, "same")
slownoise_smooth = (np.average(self.peakless) /
np.average(slownoise_smooth)) * slownoise_smooth
self.filtered_signal_1 = data_at_0 - slownoise_smooth
if ite == 0:
fig8 = plt.figure(8)
#ax8 = fig8.add_axes([0.1, 0.1, 0.8, 0.8])
#ax8.set_title('Filter Levels')
#ax8.set_xlabel('Time [$\mu$]')
#ax8.set_ylabel('Voltage [V]')
ax81 = fig8.add_subplot(311)
ax82 = fig8.add_subplot(312)
ax83 = fig8.add_subplot(313)
#ax84 = fig8.add_subplot(224)
#ax8.set_xlim(0,100)
ax81.plot(self.time, data, 'lightgrey', label='Raw Data')
ax81.plot((np.amin(self.time), np.amax(self.time)),
(meanzero, meanzero),
'b-',
label='MeanZero') #rmsn from RemovePeaksBasedOnRMS3 = 2
ax82.plot(self.time, data_at_0, 'grey', label='Raw Data Corrected')
ax82.plot(
(np.amin(self.time), np.amax(self.time)),
(meanpeakless, meanpeakless),
'g-',
label='MeanPeakLess') #rmsn from RemovePeaksBasedOnRMS3 = 2
ax83.plot(self.time, data, 'lightgrey', label='Raw Data')
ax83.plot(self.time, slownoise, 'grey', label='Slownoise')
ax83.plot((np.amin(self.time), np.amax(self.time)),
(meanzero, meanzero),
'b-',
label='MeanZero') #rmsn from RemovePeaksBasedOnRMS3 = 2
ax81.legend()
ax82.legend()
ax83.legend()
fig8.savefig(self.data_file_path + 'PeaklessGen.pdf',
format='pdf',
bbox_inches='tight')
#rawdatcorr = ax8.plot(self.time,data_at_0,'grey',label='Raw Data Corrected')
#slown = ax8.plot(self.time,slownoise,'red',label='slow Noise')
#plt.plot((np.amin(self.time),np.amax(self.time)), (level*2,level*2), 'k--',label='Cut level') #rmsn from RemovePeaksBasedOnRMS3 = 2
#plt.plot((np.amin(self.time),np.amax(self.time)), (meanzero,meanzero), 'b-',label='MeanZero') #rmsn from RemovePeaksBasedOnRMS3 = 2
#plt.plot((np.amin(self.time),np.amax(self.time)), (meanpeakless,meanpeakless), 'b--',label='MeanPeakLess') #rmsn from RemovePeaksBasedOnRMS3 = 2
#ax8.set_title('Smoothed slow Noise')
#slowns = ax8.plot(self.time,slownoise_smooth,'blue',label='Smoothed slow Noise')
#ax8.legend()
#plt.show()
#fig8.savefig(self.data_file_path+'FilterLevels.pdf',format='pdf',bbox_inches='tight')
#ax8.legend_.remove()
#fig8.clf()
self.ReportProgress(ite, self.ranger)
return self.filtered_signal_1
def gauss_filter(self, data): window = signal.general_gaussian(51, p=0.5, sig=20) filtered = signal.fftconvolve(window, data) filtered = (np.average(data) / np.average(filtered)) * filtered filtered = np.roll(filtered, -25) return filtered
def gaussian_smooth(data):
window = general_gaussian(51, p=0.5, sig=20)
filtered = fftconvolve(window, data)
filtered = (np.average(data) / np.average(filtered)) * filtered
filtered = np.roll(filtered, -25)
return filtered[0:len(data)]
# Plot the window and its frequency response:
from scipy import signal
from scipy.fftpack import fft, fftshift
import matplotlib.pyplot as plt
window = signal.general_gaussian(51, p=1.5, sig=7)
plt.plot(window)
plt.title(r"Generalized Gaussian window (p=1.5, $\sigma$=7)")
plt.ylabel("Amplitude")
plt.xlabel("Sample")
plt.figure()
A = fft(window, 2048) / (len(window) / 2.0)
freq = np.linspace(-0.5, 0.5, len(A))
response = 20 * np.log10(np.abs(fftshift(A / abs(A).max())))
plt.plot(freq, response)
plt.axis([-0.5, 0.5, -120, 0])
plt.title(r"Freq. resp. of the gen. Gaussian " r"window (p=1.5, $\sigma$=7)")
plt.ylabel("Normalized magnitude [dB]")
plt.xlabel("Normalized frequency [cycles per sample]")
def run(self):
self.info("")
self.info("SAM-FP Tools: Wavelength Calibration")
self.info("by Bruno Quint ([email protected])")
self.info("version {:s}".format(version.__str__))
self.info("Starting program.")
self.info("")
self.info("Input filename: {0.input_file:s}")
self.info("Output filename: {0.output_file:s}")
# Seistemic wavelength
try:
header = pyfits.getheader(self.input_file)
except IOError:
self.error("File not found:\n {0.input_file:s}")
self.error("Leaving program now.")
sys.exit(1)
if self.wavelength is None:
try:
self.wavelength = header['PHMWCAL']
self.info("Seistemic wavelength found in the header: "
"CRVAL = {0.wavelength:.2f} Angstrom".format(self))
except KeyError:
msg = ("No wavelength could be found. "
"Leaving program now.")
self.error(msg)
sys.exit(1)
else:
self.info("Seistemic wavelength providen by the user: "******"CRVAL = {0.wavelength:.2f} Angstrom")
# Finding the reference pixel
data = pyfits.getdata(self.input_file)
s = np.mean(data, axis=2)
s = np.mean(s, axis=1)
midpt = np.median(s)
std = np.std(s)
self.debug("Collapsed data statistics: ")
self.debug("median = {:.2f}".format(midpt))
self.debug("std = {:.2f}".format(std))
p = np.percentile(data, 50)
s_ = s.copy()
s_[s < p] = 0.
k = signal.general_gaussian(10, 1, 5)
cc = signal.correlate(s_[5:-5], k, mode="same")
self.arg_max = np.argmax(cc) + 5
self.info("Reference channel is: ")
self.info("CRPIX3 = {0.arg_max:d}")
# Find the step in angstroms
try:
z_fsr = header[self.key_zfsr]
except KeyError:
self.warn('%s card was not found in the header.' % self.key_zfsr)
z_fsr = input(' Please, enter the Free-Spectral-Range in bcv:'
'\n > ')
except TypeError:
z_fsr = input(' Please, enter the Free-Spectral-Range in bcv:'
'\n > ')
try:
z_step = header[self.key_z_step]
except KeyError:
self.warn('%s card was not found in the header.' % self.key_z_step)
z_step = input(' Please, enter the step between channels in bcv'
'\n > ')
except TypeError:
self.warn('Header was not passed to "WCal.get_wavelength_step"'
' method')
z_step = input(' Please, enter the step between channels in bcv'
'\n > ')
w_order = 2 * (self.gap_size * 1e-6) / (self.wavelength * 1e-10)
w_fsr = self.wavelength / (w_order * (1 + 1 / w_order ** 2))
queensgate_constant = self.wavelength / z_fsr
w_step = z_step / queensgate_constant
self.queesgate_constant = queensgate_constant
self.w_order = w_order
self.w_fsr = w_fsr
self.w_step = w_step
self.z_fsr = z_fsr
self.z_step = z_step
self.info('Gap size e = {0.gap_size:.1f} um')
self.info('Interference order p({0.wavelength:.02f}) = {0.w_order:.2f}')
self.info(' Z Free-Spectral-Range = {0.z_fsr:.02f} bcv')
self.info(' W Free-Spectral-Range = {0.w_fsr:.02f} A')
self.info(' Queesgate constant = {0.queesgate_constant:.02f} bcv / A')
self.info(' Z Step = {0.z_step:.03f} bcv / channel')
self.info(' W Step = {0.w_step:.04f} A / channel')
header.set('CRVAL3', value=self.wavelength,
comment='Seistemic wavelength.')
header.set('CRPIX3', value=self.arg_max)
header.set('CDELT3', value=self.w_step)
header.set('C3_3', value=self.w_step)
header.set('WCAL_W0', value=self.wavelength,
comment='Seistemic Wavelength [A]', after='PHMFIT_C')
header.set('WCAL_DW', value=self.wavelength,
comment='Wavelength increment / channel [A]',
after='PHMFIT_C')
header.add_blank('-- sam-fp wavelength calibration --',
after='PHMFIT_C')
pyfits.writeto(self.input_file.replace('.fits', '_wcal.fits'), data,
header, overwrite=True)
def get_fwhm(z, s, show=False):
"""
Returns the full-width-at-half-maximum using different models. These
models can be displayed.
Parameters
----------
z (array like) : the abscissa of the reference spectrum containing
the bcv values for each channel or the channel itself.
s (array like) : the ordinate of the reference spectrum containing
the intensity at each channel or at each bcv value.
show (bool, optional) : display plots?
Returns
-------
fwhm (float) : the full-width-at-half-maximum in units equal to z
(either channel or bcv).
"""
# Clear data
p = np.percentile(s, 50.)
s_ = s.copy()
s_[s < p] = 0.
# Find maxima avoiding the borders
k = signal.general_gaussian(10, 1, 5)
cc = signal.correlate(s_[5:-5], k, mode="same")
arg_maxima = np.array([np.argmax(cc)]) + 5
# arg_maxima = signal.argrelmax(s_[2:-2], order=5)[0]
_log.debug('Peaks found at: ' + np.array2string(arg_maxima,
separator=', '))
fitter = fitting.LevMarLSQFitter()
g_fwhm, l_fwhm, gauss = [], [], 0
for (i, argm) in enumerate(arg_maxima):
g = models.Gaussian1D(amplitude=s_[argm], mean=z[argm], stddev=3.)
g_fit = fitter(g, z[1:-1], s_[1:-1])
g_fwhm.append(g_fit.stddev * 2.355)
l_model = models.Lorentz1D(amplitude=s_[argm], x_0=z[argm], fwhm=3.)
l_fit = fitter(l_model, z[1:-1], s_[1:-1])
l_fwhm.append(l_fit.fwhm)
g_rms = np.sqrt(np.mean((s - g_fit(z)) ** 2))
l_rms = np.sqrt(np.mean((s - l_fit(z)) ** 2))
_log.info("Peak at {:2d}".format(argm))
_log.info("gaussian fit rms: {:.2f}".format(g_rms))
_log.info("lorentzian fit rms: {:.2f}".format(l_rms))
if g_rms > l_rms:
gauss += 1
else:
gauss -= 1
g_fwhm = np.mean(g_fwhm)
l_fwhm = np.mean(l_fwhm)
_log.info("Gaussian fwhm: {:.2f}".format(g_fwhm))
_log.info("Lorentzian fwhm: {:.2f}".format(l_fwhm))
if gauss > 0:
fwhm_measured = g_fwhm
elif gauss < 0:
fwhm_measured = l_fwhm
else:
fwhm_measured = (g_fwhm + l_fwhm) * 0.5
if show:
z_ = np.linspace(z[0], z[-1], 1000)
fig, axs = plt.subplots(2, 1, sharex='all')
axs[0].plot(z, s, 'ko')
axs[0].plot(z[5:-5], cc / cc.max() * s_.max(), 'y-', label='Cross-correlation')
axs[0].grid()
if len(arg_maxima) > 0:
axs[0].plot(z_, g_fit(z_), 'b-')
axs[0].plot(z_, l_fit(z_), 'r--')
axs[0].legend(loc='best')
axs[0].set_ylabel('Normalized spectrum')
for amax in arg_maxima:
axs[0].axvline(z[amax], c='k', ls='--', alpha=0.5)
# Display the errors
axs[1].plot(z, s - g_fit(z), 'bx', alpha=0.5, label='Error - Gaussian Fit')
axs[1].plot(z, s - l_fit(z), 'ro', alpha=0.5, label='Error - Lorentzian Fit')
axs[1].set_ylabel('Fir Errors [adu]')
axs[1].set_xlabel('z [bcv]')
axs[1].legend(loc='best')
axs[1].grid()
plt.tight_layout()
plt.show()
return fwhm_measured
def __init__(self, freq, spectrum):
self._freq = np.array(freq)
self._spectrum = np.array(spectrum)
self._conv_window = spsignal.general_gaussian(10, 1, 3)
def peak_finder(times, residuals, errors, transit_params, n_peaks=4,
plots=False, verbose=False, skip_priors=False):
"""
Find peaks in the residuals from a fit to a transit light curve, which
correspond to starspot occultations.
Parameters
----------
times : `numpy.ndarray`
Times [JD]
residuals : `numpy.ndarray`
Fluxes
errors : `numpy.ndarray`
Uncertainties on residuals
transit_params : `~batman.TransitParams`
Transit light curve parameters
n_peaks : bool (optional)
Number of peaks to search for. If more than `n_peaks` are found, return
only the `n_peaks` largest amplitude peaks.
plots : bool (optional)
Show diagnostic plots
verbose : bool (optional)
Warn if no peaks are found
Returns
-------
result_in_transit : list or `None`
List of all spot parameters in [amp, t0, sig, amp, t0, sig, ...] order
for spots detected.
Notes
-----
Review of minimizers tried for `peak_finder`:
`~scipy.optimize.fmin` gets amplitudes right, but doesn't vary sigmas much.
For this reason, it tends to do a better job of finding nearby, semi-
overlapping spots.
`~scipy.optimize.fmin_powell` varies amplitudes and sigmas lots, but
as a result, sometimes two nearby spots are fit with one wide gaussian.
"""
# http://stackoverflow.com/a/25666951
# Convolve residuals with a gaussian, find relative maxima
n_points_kernel = 100
window = signal.general_gaussian(n_points_kernel+1, p=1, sig=10)
filtered = signal.fftconvolve(window, residuals)
filtered = (np.max(residuals) / np.max(filtered)) * filtered
filtered = np.roll(filtered, int(-n_points_kernel/2))[:len(residuals)]
maxes = signal.argrelmax(filtered)[0]
# Only take maxima, not minima
maxes = maxes[filtered[maxes] > 0]
lower_t_bound, upper_t_bound = get_in_transit_bounds(times, transit_params)
maxes_in_transit = maxes[(times[maxes] < upper_t_bound) &
(times[maxes] > lower_t_bound)]
# Only take the `n_peaks` highest peaks
if len(maxes_in_transit) > n_peaks:
highest_maxes_in_transit = maxes_in_transit[np.argsort(filtered[maxes_in_transit])][-n_peaks:]
else:
highest_maxes_in_transit = maxes_in_transit
# plt.plot(times, filtered)
# plt.plot(times, residuals, '.')
# plt.plot(times[maxes_in_transit], filtered[maxes_in_transit], 'ro')
# [plt.axvline(times[m], color='k') for m in maxes]
# [plt.axvline(times[m], color='m') for m in maxes_in_transit]
# if len(maxes_in_transit) > n_peaks:
# [plt.axvline(times[m], color='b') for m in highest_maxes_in_transit]
# plt.axvline(upper_t_bound, color='r')
# plt.axvline(lower_t_bound, color='r')
# plt.show()
if len(maxes_in_transit) == 0:
if verbose:
print('no maxes found')
return None
peak_times = times[highest_maxes_in_transit]
peak_amplitudes = residuals[highest_maxes_in_transit]
peak_sigmas = np.zeros(len(peak_times)) + 2./60/24 # 3 min
input_parameters = np.vstack([peak_amplitudes, peak_times,
peak_sigmas]).T.ravel()
result = optimize.fmin_powell(peak_finder_chi2, input_parameters,
disp=False, args=(times, residuals, errors),
xtol=0.00001, ftol=0.00001)
# if np.all(result == input_parameters):
# print('oh no!, fmin didnt produce a fit')
# Only use gaussians that occur in transit (fmin fit is unbounded in time)
# and amplitude is positive:
split_result = np.split(result, len(input_parameters)/3)
result_in_transit = []
for amplitude, t0, sigma in split_result:
depth = transit_params.rp**2
trial_params = np.array([amplitude, t0, sigma])
if not np.isinf(lnprior(trial_params, residuals, lower_t_bound,
upper_t_bound, transit_params, skip_priors)):
result_in_transit.extend([amplitude, t0, np.abs(sigma)])
result_in_transit = np.array(result_in_transit)
if len(result_in_transit) == 0:
return None
if plots:
fig, ax = plt.subplots(3, 1, figsize=(8, 10), sharex=True)
ax[0].errorbar(times, residuals, fmt='.', color='k')
[ax[0].axvline(t) for t in result_in_transit[1::3]]
ax[0].plot(times, summed_gaussians(times, input_parameters), 'r')
ax[0].axhline(0, color='k', ls='--')
ax[0].set_ylabel('Transit Residuals')
ax[1].errorbar(times, residuals, fmt='.', color='k')
ax[1].plot(times, summed_gaussians(times, result_in_transit), 'r')
ax[1].axhline(0, color='k', ls='--')
ax[1].set_ylabel('Residuals')
ax[2].errorbar(times,
residuals - summed_gaussians(times, result_in_transit),
fmt='.', color='k')
#ax[1].errorbar(times, gaussian_model, fmt='.', color='r')
ax[2].axhline(0, color='k', ls='--')
ax[2].set_ylabel('Residuals')
for axis in ax:
axis.axvline(upper_t_bound, color='r')
axis.axvline(lower_t_bound, color='r')
fig.tight_layout()
plt.show()
return result_in_transit
# Plot the window and its frequency response:
from scipy import signal
from scipy.fftpack import fft, fftshift
import matplotlib.pyplot as plt
window = signal.general_gaussian(51, p=1.5, sig=7)
plt.plot(window)
plt.title(r"Generalized Gaussian window (p=1.5, $\sigma$=7)")
plt.ylabel("Amplitude")
plt.xlabel("Sample")
plt.figure()
A = fft(window, 2048) / (len(window)/2.0)
freq = np.linspace(-0.5, 0.5, len(A))
response = 20 * np.log10(np.abs(fftshift(A / abs(A).max())))
plt.plot(freq, response)
plt.axis([-0.5, 0.5, -120, 0])
plt.title(r"Freq. resp. of the gen. Gaussian window (p=1.5, $\sigma$=7)")
plt.ylabel("Normalized magnitude [dB]")
plt.xlabel("Normalized frequency [cycles per sample]")
def resconvolve(sourceimage, lowres, highres, resultimage = 'same'):
"""
Convolve sourceimage to a lower resolution image.
Parameters
----------
sourceimage : string
Rath to image to be convolved.
lowres : float
Resolution of the current image (in arcsec).
highres : float
Resolution of the requied convolved image (in arcsec).
resultimage : string
If 'same', then use the same directoy and file name is sourceimage.resamp.fits.
If specified then use the given path and file name.
Default is 'same'.
Returns
-------
saves a file : fits
Stores an 2-D image of sourceimage convolved to the required resolution at resultimage.
The header of sourceimage is used.
"""
# Load in source data and header information
data,header = fits.getdata(sourceimage,header = True)
# nan to zero
data[np.isnan(data)] = 0
# number of x and y pixels
x = header['NAXIS2']
y = header['NAXIS1']
# get the pixelsize in arcsec, where the value is in degree*3600(arcsec/degree)
pixelsize = header['CDELT2']*3600
# gaussian consant to convert sigma to FWHM 2*sqrt(2ln(2))
constant = 2*np.sqrt(2*np.log(2))
# FWHM of current image
FWHM_high = highres/pixelsize
# sigma of current image
sigma_high = FWHM_high/constant
# FWHM of required resolution
FWHM_low = lowres/pixelsize
# sigma of required resolution
sigma_low = FWHM_low/constant
# FWHM calulated for the gaussian of convolution kernel
FWHM = np.sqrt(FWHM_low**2 - FWHM_high**2)
# sigma for the gaussian of convolution kernel
sigma = FWHM/constant
# making the 2-D image of the convolution kernel by making 2, 1-D gaussian
# and normalized by the gaussian normalization factor
gauss1 = signal.general_gaussian(x,1, sigma)/((sigma)*np.sqrt(2*np.pi))
gauss2 = signal.general_gaussian(y,1, sigma)/((sigma)*np.sqrt(2*np.pi))
# combine these two to create 2D image
kernel = np.outer(gauss1, gauss2)
# convolve the image using signal.fftconvolve premade function and kernel
convolved = signal.fftconvolve(data, kernel, mode='same')
# change the resolution if it's there
header['RESO'] = (lowres, '[arcsec] Resolution of the map')
# name of the path based on the resultimage
if resultimage == 'same':
path = sourceimage[:-5]+'.conv.fits'
else:
path = resultimage
# saves as the fits file
fits.writeto(path, convolved, header)
print "Convolving done. File is saved. \n"
return
