inspiral_extrinsics.py: Performance improvments to dt likelihood ratio code

gstlal-inspiral/python/stats/inspiral_extrinsics.py

@@ -699,6 +699,19 @@ class SNRPDF(object):
 #
 
 
+dt_chebyshev_coeffs_polynomials = []
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([-597.94227757949329, 2132.773853473127, -2944.126306979932, 1945.3033368083029, -603.91576991927593, 70.322754873993347]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([-187.50681052710425, 695.84172327044325, -1021.0688423797938, 744.3266490236075, -272.12853221391498, 35.542404632554508]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([52.128579054466599, -198.32054234352267, 301.34865541080791, -230.8522943433488, 90.780611645135437, -16.310130528927655]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([48.216566165393878, -171.70632176976451, 238.48370471322843, -159.65005032451785, 50.122296925755677, -5.5466740894321367]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([-34.030336093450863, 121.44714070928059, -169.91439486329773, 115.86873916702386, -38.08411813067778, 4.7396784315927816]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([3.4576823675810178, -12.362609260376738, 17.3300203922424, -11.830868787176165, 3.900284020272442, -0.47157315012399248]))
+dt_chebyshev_coeffs_polynomials.append(numpy.poly1d([4.0423239651315113, -14.492611904657275, 20.847419746265583, -15.033846689362553, 5.3953159232942216, -0.78132676885883601]))
+norm_polynomial = numpy.poly1d([-348550.84040194791, 2288151.9147818103, -6623881.5646601757, 11116243.157047395, -11958335.1384027, 8606013.1361163966, -4193136.6690072878, 1365634.0450674745, -284615.52077054407, 34296.855844416605, -1815.7135263788341])
+
+dt_chebyshev_coeffs = [0]*13
+
+
 def coinc_window(delta_t, instruments):
 	if len(instruments) == 1:
 		return 0.
@@ -751,18 +764,16 @@ def lnP_dt_signal(dt, snr_ratio):
 	# Fits below an snr ratio of 0.33 are not reliable
 	if snr_ratio < 0.33:
 		snr_ratio = 0.33
-	x = numpy.zeros(13)
-
-	x[0] = numpy.poly1d([-597.94227757949329, 2132.773853473127, -2944.126306979932, 1945.3033368083029, -603.91576991927593, 70.322754873993347])(snr_ratio)
-	x[2] = numpy.poly1d([-187.50681052710425, 695.84172327044325, -1021.0688423797938, 744.3266490236075, -272.12853221391498, 35.542404632554508])(snr_ratio)
-	x[4] = numpy.poly1d([52.128579054466599, -198.32054234352267, 301.34865541080791, -230.8522943433488, 90.780611645135437, -16.310130528927655])(snr_ratio)
-	x[6] = numpy.poly1d([48.216566165393878, -171.70632176976451, 238.48370471322843, -159.65005032451785, 50.122296925755677, -5.5466740894321367])(snr_ratio)
-	x[8] = numpy.poly1d([-34.030336093450863, 121.44714070928059, -169.91439486329773, 115.86873916702386, -38.08411813067778, 4.7396784315927816])(snr_ratio)
-	x[10] = numpy.poly1d([3.4576823675810178, -12.362609260376738, 17.3300203922424, -11.830868787176165, 3.900284020272442, -0.47157315012399248])(snr_ratio)
-	x[12] = numpy.poly1d([4.0423239651315113, -14.492611904657275, 20.847419746265583, -15.033846689362553, 5.3953159232942216, -0.78132676885883601])(snr_ratio)
-	norm = numpy.poly1d([-348550.84040194791, 2288151.9147818103, -6623881.5646601757, 11116243.157047395, -11958335.1384027, 8606013.1361163966, -4193136.6690072878, 1365634.0450674745, -284615.52077054407, 34296.855844416605, -1815.7135263788341])(snr_ratio)
-
-	return numpy.polynomial.chebyshev.chebval(dt/0.015013, x) - numpy.log(norm)
+
+	dt_chebyshev_coeffs[0] = dt_chebyshev_coeffs_polynomials[0](snr_ratio)
+	dt_chebyshev_coeffs[2] = dt_chebyshev_coeffs_polynomials[1](snr_ratio)
+	dt_chebyshev_coeffs[4] = dt_chebyshev_coeffs_polynomials[2](snr_ratio)
+	dt_chebyshev_coeffs[6] = dt_chebyshev_coeffs_polynomials[3](snr_ratio)
+	dt_chebyshev_coeffs[8] = dt_chebyshev_coeffs_polynomials[4](snr_ratio)
+	dt_chebyshev_coeffs[10] = dt_chebyshev_coeffs_polynomials[5](snr_ratio)
+	dt_chebyshev_coeffs[12] = dt_chebyshev_coeffs_polynomials[6](snr_ratio)
+
+	return numpy.polynomial.chebyshev.chebval(dt/0.015013, dt_chebyshev_coeffs) - numpy.log(norm_polynomial(snr_ratio))
 
 def lnP_dt_dphi_uniform_H1L1(coincidence_window_extension):
 	# FIXME Dont hardcode