Revert "far.py: don't convolve snr,chi numerators with KDE"

This reverts commit fef0db69.

gstlal-inspiral/python/far.py

@@ -934,16 +934,10 @@ class ThincaCoincParamsDistributions(snglcoinc.CoincParamsDistributions):
 					progressbar.increment()
 				new_binarr.array[snrindices, rcossindices] += ncx2pdf(snrchi2, df, numpy.array([pf * snrs2]).T)
 
-			# apply a bit of smoothing (5 bins in each
-			# direction) to soften the stair steps caused by
-			# the limited set of discrete prefactors
-			rate.filter_array(new_binarr.array, rate.gaussian_window(5., 5., sigma = 10.))
-
 			# Add an SNR power law in with the differentials
 			dsnrdchi2 = numpy.outer(dsnr / snr**inv_snr_pow, drcoss)
 			new_binarr.array[snrindices, rcossindices] *= dsnrdchi2
 			new_binarr.array[snrindices, rcossindices] *= number_of_events / new_binarr.array.sum()
-
 			# add to raw counts
 			binarr += new_binarr
 
@@ -1130,38 +1124,31 @@ class ThincaCoincParamsDistributions(snglcoinc.CoincParamsDistributions):
 	def pdf_from_rates_snrchi2(self, key, pdf_dict, snr_kernel_width_at_8 = 10., chisq_kernel_width = 0.1,  sigma = 10.):
 		# get the binned array we're going to process
 		binnedarray = pdf_dict[key]
-
-		# if this is not the numerator, convolve with density
-		# estimation kernel (numerator is known analytically).
-		if pdf_dict is not self.injection_pdf:
-			# total number of samples.  be conservative and
-			# assume only 1 in 10 samples are independent.
-			numsamples = binnedarray.array.sum() / 10. + 1.
-
-			# construct the density estimation kernel.  apply
-			# Silverman's rule so that the width scales with
-			# numsamples**(-1./6.) for a 2D PDF.  don't let the
-			# kernel get smaller than the 2.5 times the bin
-			# size.
-			snr_bins = binnedarray.bins[0]
-			chisq_bins = binnedarray.bins[1]
-			snr_per_bin_at_8 = (snr_bins.upper() - snr_bins.lower())[snr_bins[8.]]
-			chisq_per_bin_at_0_02 = (chisq_bins.upper() - chisq_bins.lower())[chisq_bins[0.02]]
-
-			snr_kernel_bins = snr_kernel_width_at_8 / snr_per_bin_at_8 / numsamples**(1./6.)
-			chisq_kernel_bins = chisq_kernel_width / chisq_per_bin_at_0_02 / numsamples**(1./6.)
-
-			if  snr_kernel_bins < 2.5:
-				snr_kernel_bins = 2.5
-				warnings.warn("Replacing snr kernel bins with 2.5")
-			if  chisq_kernel_bins < 2.5:
-				chisq_kernel_bins = 2.5
-				warnings.warn("Replacing chisq kernel bins with 2.5")
-
-			kernel = rate.gaussian_window(snr_kernel_bins, chisq_kernel_bins, sigma = sigma)
-
-			# convolve with the bin count data
-			rate.filter_array(binnedarray.array, kernel)
+		numsamples = binnedarray.array.sum() / 10. + 1. # Be extremely conservative and assume only 1 in 10 samples are independent.
+		# construct the density estimation kernel
+		snr_bins = binnedarray.bins[0]
+		chisq_bins = binnedarray.bins[1]
+		snr_per_bin_at_8 = (snr_bins.upper() - snr_bins.lower())[snr_bins[8.]]
+		chisq_per_bin_at_0_02 = (chisq_bins.upper() - chisq_bins.lower())[chisq_bins[0.02]]
+
+		# Apply Silverman's rule so that the width scales with numsamples**(-1./6.) for a 2D PDF
+		snr_kernel_bins = snr_kernel_width_at_8 / snr_per_bin_at_8 / numsamples**(1./6.)
+		chisq_kernel_bins = chisq_kernel_width / chisq_per_bin_at_0_02 / numsamples**(1./6.)
+
+		# check the size of the kernel. We don't ever let it get
+		# smaller than the 2.5 times the bin size
+		if  snr_kernel_bins < 2.5:
+			snr_kernel_bins = 2.5
+			warnings.warn("Replacing snr kernel bins with 2.5")
+		if  chisq_kernel_bins < 2.5:
+			chisq_kernel_bins = 2.5
+			warnings.warn("Replacing chisq kernel bins with 2.5")
+
+		# Compute the KDE
+		kernel = rate.gaussian_window(snr_kernel_bins, chisq_kernel_bins, sigma = sigma)
+
+		# convolve with the bin count data
+		rate.filter_array(binnedarray.array, kernel)
 
 		# zero everything below the SNR cut-off.  need to do the
 		# slicing ourselves to avoid zeroing the at-threshold bin