inspiral_extrinsics: increase marginalization accuracy in dt / dphi code and broaden the covariance

gstlal-inspiral/python/stats/inspiral_extrinsics.py

@@ -1027,9 +1027,9 @@ def margprob(Dmat):
 	out = []
 	for D in Dmat:
 		D = D[numpy.isfinite(D)]
-		step = max(int(len(D) / 2048.), 1)
+		step = max(int(len(D) / 65536.), 1)
 		D = D[::step]
-		if len(D) == 2049:
+		if len(D) == 65537:
 			out.append(step * scipy.integrate.romb(numpy.exp(-D**2/2.)))
 		else:
 			out.append(step * scipy.integrate.simps(numpy.exp(-D**2/2.)))
@@ -1145,14 +1145,14 @@ class TimePhaseSNR(object):
 	numchunks = 20
 
 	# These are generated by running
-	# ./gstlal_inspiral_compute_dtdphideff_cov_matrix --psd-xml 2019-05-09-H1L1V1psd_new.xml.gz --H-snr 5.0 --L-snr 7.0 --V-snr 2.25
+	# ./gstlal_inspiral_compute_dtdphideff_cov_matrix --psd-xml gstlal-inspiral/share/O3/2019-05-09-H1L1V1psd_new.xml.gz --H-snr 4.3 --L-snr 6.0 --V-snr 2.
 	# NOTE NOTE NOTE: You cannot just change these without fully rebuilding
 	# the trees and the marginalized distributions.
-	transtt = {frozenset(("H1", "L1")): 5.19313416e+04, frozenset(("H1", "V1")): 1.39663777e+04, frozenset(("L1", "V1")): 3.98320074e+04}
-	transpp = {frozenset(("H1", "L1")): 3.50152221e+00, frozenset(("H1", "V1")): 1.56080114e+00, frozenset(("L1", "V1")): 1.63248925e+00}
-	transtp = {frozenset(("H1", "L1")): -6.51410977e+01, frozenset(("H1", "V1")): -2.07581116e+01, frozenset(("L1", "V1")): -6.05964763e+01}
-	transpt = {frozenset(("H1", "L1")): 0.0, frozenset(("H1", "V1")): 0.0, frozenset(("L1", "V1")): 0.0}
-	transdd = {frozenset(("H1", "L1")): 5.0 , frozenset(("H1", "V1")): 5.0, frozenset(("L1", "V1")): 5.0}
+	transtt = {frozenset(['V1', 'H1']): 12069.320302753127, frozenset(['H1', 'L1']): 44554.743732089875, frozenset(['V1', 'L1']): 34221.089108680397}
+	transtp = {frozenset(['V1', 'H1']): -17.906651775391733, frozenset(['H1', 'L1']): -55.897144921848493, frozenset(['V1', 'L1']): -52.035238588072133}
+	transpt = {frozenset(['V1', 'H1']): 0.0, frozenset(['H1', 'L1']): 0.0, frozenset(['V1', 'L1']): 0.0}
+	transpp = {frozenset(['V1', 'H1']): 1.3806845740864522, frozenset(['H1', 'L1']): 3.0084592948751365, frozenset(['V1', 'L1']): 1.4477360415175331}
+	transdd = {frozenset(['V1', 'H1']): 5.0, frozenset(['H1', 'L1']): 5.0, frozenset(['V1', 'L1']): 5.0}
 
 	def __init__(self, tree_data = None, margsky = None, verbose = False, margstart = 0, margstop = None):
 		"""
@@ -1236,7 +1236,7 @@ class TimePhaseSNR(object):
 				for cnt, points in enumerate(chunker(self.tree_data[:,slcs], self.numchunks, margstart, margstop)):
 					if verbose:
 						print >> sys.stderr, "%d/%d" % (cnt * self.numchunks, num_points)
-					Dmat = self.KDTree[combo].query(points, k=num_points, distance_upper_bound = 8.5)[0]
+					Dmat = self.KDTree[combo].query(points, k=num_points, distance_upper_bound = 15.0)[0]
 					marg[margstart + self.numchunks * cnt : margstart + self.numchunks * (cnt+1)] = margprob(Dmat)
 				self.margsky[combo] = numpy.array(marg, dtype="float32")