Maintenance will be performed on git.ligo.org, chat.ligo.org, containers.ligo.org, and docs.ligo.org on Tuesday 7th July 2020 starting at approximately 10am PDT and lasting for around 15 minutes. There will be a short period of downtime towards the end of the maintenance window. Please direct any comments, questions, or concerns to uwm-help@cgca.uwm.edu.

Commit 351d1ad8 authored by Chad Hanna's avatar Chad Hanna Committed by CHAD RICHARD HANNA

inspiral_extrinsics.py, inspiral_lr.py: add glitch model rather than repurposing the signal model

parent f8429e6e
Pipeline #75887 passed with stages
in 26 minutes and 58 seconds
......@@ -963,6 +963,47 @@ class NumeratorSNRCHIPDF(rate.BinnedLnPDF):
# add to lnpdf
lnpdf.array += arr
@staticmethod
def add_glitch_model(lnpdf, n, prefactors_range, df, inv_snr_pow = 4., snr_min = 3.5, progressbar = None):
if df <= 0.:
raise ValueError("require df >= 0: %s" % repr(df))
pfs = numpy.linspace(prefactors_range[0], prefactors_range[1], 100)
if progressbar is not None:
progressbar.max = len(pfs)
# FIXME: except for the low-SNR cut, the slicing is done
# to work around various overflow and loss-of-precision
# issues in the extreme parts of the domain of definition.
# it would be nice to identify the causes of these and
# either fix them or ignore them one-by-one with a comment
# explaining why it's OK to ignore the ones being ignored.
# for example, computing snrchi2 by exponentiating the sum
# of the logs of the terms might permit its evaluation
# everywhere on the domain. can ncx2pdf() be made to work
# everywhere?
snrindices, rcossindices = lnpdf.bins[snr_min:1e10, 1e-10:1e10]
snr, dsnr = lnpdf.bins[0].centres()[snrindices], lnpdf.bins[0].upper()[snrindices] - lnpdf.bins[0].lower()[snrindices]
rcoss, drcoss = lnpdf.bins[1].centres()[rcossindices], lnpdf.bins[1].upper()[rcossindices] - lnpdf.bins[1].lower()[rcossindices]
snr2 = snr**2.
snrchi2 = numpy.outer(snr2, rcoss) * df
arr = numpy.zeros_like(lnpdf.array)
for pf in pfs:
if progressbar is not None:
progressbar.increment()
arr[snrindices, rcossindices] += gstlalstats.ncx2pdf(snrchi2, df, numpy.array([pf * snr2]).T)
# convert to counts by multiplying by bin volume, and also
# multiply by an SNR powr law
arr[snrindices, rcossindices] *= numpy.outer(dsnr / snr**inv_snr_pow, drcoss)
# normalize to a total count of n
arr *= n / arr.sum()
# add to lnpdf
lnpdf.array += arr
def to_xml(self, *args, **kwargs):
elem = super(rate.BinnedLnPDF, self).to_xml(*args, **kwargs)
elem.appendChild(ligolw_array.Array.build("norm", self.norm))
......
......@@ -831,7 +831,7 @@ class LnNoiseDensity(LnLRDensity):
# add in the 99% noise model
lnpdf.array += arr
# add 1% from the "glitch model"
inspiral_extrinsics.NumeratorSNRCHIPDF.add_signal_model(lnpdf, n = 0.01 * number_of_events, prefactors_range = prefactors_range, df = df, inv_snr_pow = inv_snr_pow, snr_min = self.snr_min)
inspiral_extrinsics.NumeratorSNRCHIPDF.add_glitch_model(lnpdf, n = 0.01 * number_of_events, prefactors_range = prefactors_range, df = df, inv_snr_pow = inv_snr_pow, snr_min = self.snr_min)
# re-normalize
lnpdf.normalize()
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment