OEM pdf health fix

The sum of noise counts may change due to first-round-extinction, so this commit stores the sum of counts in the PDF before extinction in a different variable, and this variable is used for checking the health of the PDF (which is exactly what was being done before)

gstlal-inspiral/python/far.py

@@ -645,6 +645,11 @@ class RankingStatPDF(object):
 		self.signal_lr_lnpdf.array *= rankingstat.numerator.candidate_count_model() / self.signal_lr_lnpdf.array.sum()
 		self.signal_lr_lnpdf.normalize()
 
+		# the sum of noise counts may change due to first-round-extinction,
+		# so store the sum of counts before extinction in a different variable.
+		# This will be used for checking the health of the PDF
+		self.noise_counts_before_extinction = self.noise_lr_lnpdf.array.sum()
+
 
 	def copy(self):
 		new = self.__class__(None)
@@ -653,6 +658,7 @@ class RankingStatPDF(object):
 		new.zero_lag_lr_lnpdf = self.zero_lag_lr_lnpdf.copy()
 		new.segments = type(self.segments)(self.segments)
 		new.template_ids = self.template_ids
+		new.noise_counts_before_extinction = self.noise_counts_before_extinction
 		return new
 
 
@@ -719,6 +725,7 @@ WHERE
 		# templates are the same then the segments must be disjoint
 		self.segments += other.segments
 		self.template_ids |= other.template_ids
+		self.noise_counts_before_extinction += other.noise_counts_before_extinction
 		return self
 
 
@@ -796,7 +803,7 @@ WHERE
 	def is_healthy(self, verbose = False):
 		# do we believe the PDFs are sufficiently well-defined to
 		# compute FAPs and FARs?
-		health = min(self.noise_lr_lnpdf.array.sum() / 1000000., self.zero_lag_lr_lnpdf.array.sum() / 1000.)
+		health = min(self.noise_counts_before_extinction / 1000000., self.zero_lag_lr_lnpdf.array.sum() / 1000.)
 		with tqdm(desc = "ranking stat. health", total = 1., bar_format="{l_bar}{bar}", disable = not verbose) as pbar:
 			pbar.update(health)
 		return health >= 1.
@@ -830,6 +837,13 @@ WHERE
 		self.segments = ligolw_param.get_pyvalue(xml, u"segments").strip()
 		self.segments = segmentsUtils.from_range_strings(self.segments.split(",") if self.segments else [], float)
 		self.template_ids = frozenset(map(int, ligolw_param.get_pyvalue(xml, u"template_ids").split(",")))
+		try:
+			self.noise_counts_before_extinction = ligolw_param.get_pyvalue(xml, u"noise_counts_before_extinction")
+		except ValueError:
+			# the PDF was using the old extinction model
+			# in which case noise_counts_before_extinction
+			# is the same as the sum of the noise array
+			self.noise_counts_before_extinction = self.noise_lr_lnpdf.array.sum()
 		return self
 
 	def to_xml(self, name):
@@ -845,6 +859,7 @@ WHERE
 		xml.appendChild(self.zero_lag_lr_lnpdf.to_xml(u"zero_lag_lr_lnpdf"))
 		xml.appendChild(ligolw_param.Param.from_pyvalue(u"segments", ",".join(segmentsUtils.to_range_strings(self.segments))))
 		xml.appendChild(ligolw_param.Param.from_pyvalue(u"template_ids", ",".join("%d" % template_id for template_id in sorted(self.template_ids))))
+		xml.appendChild(ligolw_param.Param.from_pyvalue(u"noise_counts_before_extinction", int(self.noise_counts_before_extinction)))
 		return xml