def check_input(times,signal,**kwargs):
"""
Check the input arguments for periodogram calculations for mistakes.
If you get an error when trying to compute a periodogram, and you don't
understand it, just feed the input you gave to this function, and it will
perform some basic checks.
"""
#-- check if the input are arrays and have the same 1D shape
is_array0 = isinstance(times,np.ndarray)
is_array1 = isinstance(signal,np.ndarray)
if not is_array0: print(termtools.red('ERROR: time input is not an array'))
if not is_array1: print(termtools.red('ERROR: signal input is not an array'))
if not is_array0 or not is_array1:
times = np.asarray(times)
signal = np.asarray(signal)
print(termtools.green("---> FIXED: inputs are arrays"))
print(termtools.green("OK: inputs are arrays"))
onedim = (len(times.shape)==1) & (len(signal.shape)==1)
same_shape = times.shape==signal.shape
if not onedim or not same_shape:
print(termtools.red('ERROR: input is not 1D or not of same length'))
return False
print(termtools.green("OK: inputs are 1D and have same length"))
#-- check if the signal constains nans or infs:
isnan0 = np.sum(np.isnan(times))
isnan1 = np.sum(np.isnan(signal))
isinf0 = np.sum(np.isinf(times))
isinf1 = np.sum(np.isinf(signal))
if isnan0: print(termtools.red('ERROR: time array contains nans'))
if isnan1: print(termtools.red('ERROR: signal array contains nans'))
if isinf0: print(termtools.red('ERROR: time array contains infs'))
if isinf1: print(termtools.red('ERROR: signal array contains infs'))
if not isnan0 and not isnan1 and not isinf0 and not isinf1:
print(termtools.green('OK: no infs or nans'))
else:
keep = -np.isnan(times) & -np.isnan(signal) & -np.isinf(times) & -np.isinf(signal)
times,signal = times[keep],signal[keep]
print(termtools.green('---> FIXED: infs and nans removed'))
#-- check if the timeseries is sorted
is_sorted = np.all(np.diff(times)>0)
if not is_sorted:
print(termtools.red('ERROR: time array is not sorted'))
sa = np.argsort(times)
times,signal = times[sa],signal[sa]
print(termtools.green('---> FIXED: time array is sorted'))
else:
print(termtools.green("OK: time array is sorted"))
print(termtools.green("No inconsistencies found or inconsistencies are fixed"))
#-- check keyword arguments:
fnyq = getNyquist(times,nyq_stat=np.min)
print("Default Nyquist frequency: {}".format(fnyq))
if 'nyq_stat' in kwargs:
fnyq = getNyquist(times,nyq_stat=kwargs['nyq_stat'])
print("Nyquist value manually set to {}".format(fnyq))
if 'fn' in kwargs and kwargs['fn']>fnyq:
print(termtools.red("Final frequency 'fn' is larger than the Nyquist frequency"))
return times,signal
def scargle_probability(peak_value,
times,
freqs,
correct_for_frange=False,
**kwargs):
"""
Compute the probability to observe a peak in the Scargle periodogram.
If C{correct_for_frange=True}, the Bonferroni correction will be applied
to a smaller number of frequencies (i.e. the independent number of
frequencies in C{freqs}). To be conservative, set C{correct_for_frange=False}.
Example simulation:
>>> times = np.linspace(0,1,5000)
>>> N = 500
>>> probs = np.zeros(N)
>>> peaks = np.zeros(N)
>>> for i in range(N):
... signal = np.random.normal(size=len(times))
... f,s = scargle(times,signal,threads='max',norm='distribution')
... peaks[i] = s.max()
... probs[i] = scargle_probability(s.max(),times,f)
Now make a plot:
>>> p = pl.figure()
>>> p = pl.subplot(131)
>>> p = pl.plot(probs,'ko')
>>> p = pl.plot([0,N],[0.01,0.01],'r-',lw=2)
>>> p = pl.subplot(132)
>>> p = pl.plot(peaks[np.argsort(peaks)],probs[np.argsort(peaks)],'ro')
>>> p = pl.plot(peaks[np.argsort(peaks)],1-(1-np.exp(-np.sort(peaks)))**(10000.),'g-')
>>> #p = pl.plot(peaks[np.argsort(peaks)],1-(1-(1-np.sort(peaks)/2500.)**2500.)**(10000.),'b--')
>>> p = pl.subplot(133)
>>> for i in np.logspace(-3,0,100):
... p = pl.plot([i*100],[np.sum(probs<i)/float(N)*100],'ko')
>>> p = pl.plot([1e-6,100],[1e-6,100],'r-',lw=2)
>>> p = pl.xlabel('Should observe this many points below threshold')
>>> p = pl.ylabel('Observed this many points below threshold')
]]include figure]]ivs_timeseries_pergrams_prob.png]
"""
#-- independent frequencies
nr_obs = len(times)
ni = 2 * nr_obs
#-- correct the nr of independent frequencies for the frequency range
# that is tested, but only if it is requested
if correct_for_frange:
nyqstat = kwargs.pop('nyqstat', np.min)
nyquist = getNyquist(times, nyqstat=nyqstat)
ni = int(freqs.ptp() / nyquist * ni)
#p_value = 1. - (1.- (1-2*peak_value/nr_obs)**(nr_obs/2))**ni
p_value = 1. - (1. - np.exp(-peak_value))**ni
return p_value
def scargle_probability(peak_value,times,freqs,correct_for_frange=False,**kwargs):
"""
Compute the probability to observe a peak in the Scargle periodogram.
If C{correct_for_frange=True}, the Bonferroni correction will be applied
to a smaller number of frequencies (i.e. the independent number of
frequencies in C{freqs}). To be conservative, set C{correct_for_frange=False}.
Example simulation:
>>> times = np.linspace(0,1,5000)
>>> N = 500
>>> probs = np.zeros(N)
>>> peaks = np.zeros(N)
>>> for i in range(N):
... signal = np.random.normal(size=len(times))
... f,s = scargle(times,signal,threads='max',norm='distribution')
... peaks[i] = s.max()
... probs[i] = scargle_probability(s.max(),times,f)
Now make a plot:
>>> p = pl.figure()
>>> p = pl.subplot(131)
>>> p = pl.plot(probs,'ko')
>>> p = pl.plot([0,N],[0.01,0.01],'r-',lw=2)
>>> p = pl.subplot(132)
>>> p = pl.plot(peaks[np.argsort(peaks)],probs[np.argsort(peaks)],'ro')
>>> p = pl.plot(peaks[np.argsort(peaks)],1-(1-np.exp(-np.sort(peaks)))**(10000.),'g-')
>>> #p = pl.plot(peaks[np.argsort(peaks)],1-(1-(1-np.sort(peaks)/2500.)**2500.)**(10000.),'b--')
>>> p = pl.subplot(133)
>>> for i in np.logspace(-3,0,100):
... p = pl.plot([i*100],[np.sum(probs<i)/float(N)*100],'ko')
>>> p = pl.plot([1e-6,100],[1e-6,100],'r-',lw=2)
>>> p = pl.xlabel('Should observe this many points below threshold')
>>> p = pl.ylabel('Observed this many points below threshold')
]]include figure]]ivs_timeseries_pergrams_prob.png]
"""
#-- independent frequencies
nr_obs = len(times)
ni = 2*nr_obs
#-- correct the nr of independent frequencies for the frequency range
# that is tested, but only if it is requested
if correct_for_frange:
nyqstat = kwargs.pop('nyqstat',np.min)
nyquist = getNyquist(times,nyqstat=nyqstat)
ni = int(freqs.ptp()/nyquist*ni)
#p_value = 1. - (1.- (1-2*peak_value/nr_obs)**(nr_obs/2))**ni
p_value = 1. - (1.- np.exp(-peak_value))**ni
return p_value
