def calculate_ifar(self, newsnr, duration):
if self.fit_info['fixed_ifar']:
return self.fit_info['fixed_ifar']
dur_bin = self.fit_info['bins'][duration]
rate = self.fit_info['rates'][dur_bin]
coeff = self.fit_info['coeffs'][dur_bin]
rate_louder = rate * fits.cum_fit('exponential', [newsnr], coeff,
self.fit_info['thresh'])[0]
# apply a trials factor of the number of duration bins
rate_louder *= len(self.fit_info['rates'])
return conv.sec_to_year(1. / rate_louder)
def calculate_ifar(self, sngl_ranking, duration):
if self.fixed_ifar:
return self.fixed_ifar[self.ifo]
with h5py.File(self.fit_file, 'r') as fit_file:
bin_edges = fit_file['bins_edges'][:]
live_time = fit_file[self.ifo].attrs['live_time']
thresh = fit_file.attrs['fit_threshold']
dist_grp = fit_file[self.ifo][self.sngl_ifar_est_dist]
rates = dist_grp['counts'][:] / live_time
coeffs = dist_grp['fit_coeff'][:]
bins = bin_utils.IrregularBins(bin_edges)
dur_bin = bins[duration]
rate = rates[dur_bin]
coeff = coeffs[dur_bin]
rate_louder = rate * fits.cum_fit('exponential', [sngl_ranking], coeff,
thresh)[0]
# apply a trials factor of the number of duration bins
rate_louder *= len(rates)
return conv.sec_to_year(1. / rate_louder)
def n_louder_from_fit(back_stat,
fore_stat,
dec_facs,
fit_func='exponential',
fit_thresh=0):
"""
Use a fit to events in back_stat in order to estimate the
distribution for use in recovering the estimate count of louder
background events. Below the fit threshold, use the n_louder
method for these triggers
back_stat: numpy.ndarray
Array of the background statistic values
fore_stat: numpy.ndarray or scalar
Array of the foreground statistic values or single value
dec_facs: numpy.ndarray
Array of the decimation factors for the background statistics
fit_func: str
Name of the function to be used for the fit to background
statistic values
fit_thresh: float
Threshold above which triggers use the fitted value, below this
the counted number of louder events will be used
Returns
-------
back_cnum: numpy.ndarray
The estimated number of background events louder than each
background event
fn_louder: numpy.ndarray
The estimated number of background events louder than each
foreground event
"""
# Calculate the fitting factor of the ranking statistic distribution
alpha, _ = trstats.fit_above_thresh(fit_func,
back_stat,
thresh=fit_thresh,
weights=dec_facs)
# Count background events above threshold as the cum_fit is
# normalised to 1
bg_above = back_stat > fit_thresh
bg_above_thresh = np.sum(dec_facs[bg_above])
fg_above = fore_stat > fit_thresh
# These will be overwritten, but just to silence a warning
# in the case where trstats.cum_fit returns zero
back_cnum = np.zeros_like(back_stat)
fnlouder = np.zeros_like(fore_stat)
# Ue the fit above the threshold
back_cnum[bg_above] = trstats.cum_fit(fit_func, back_stat[bg_above], alpha,
fit_thresh) * bg_above_thresh
fnlouder[fg_above] = trstats.cum_fit(fit_func, fore_stat[fg_above], alpha,
fit_thresh) * bg_above_thresh
# Below the fit threshold, we expect there to be sufficient events
# to use the count_n_louder method, and the distribution may deviate
# from the fit function
fg_below = np.logical_not(fg_above)
bg_below = np.logical_not(bg_above)
back_cnum[bg_below], fnlouder[fg_below] = \
count_n_louder(back_stat, fore_stat, dec_facs)
return back_cnum, fnlouder