Right now, our alpha computation is based on a truth curve created by resampling a much larger NFW truth set to match the redshift distribution of our sample. @mcclearyj has shown this to result in some shot noise in alpha of order a few percent, so we should instead take the mean of multiple subsamples in the shear-profiles module to eliminate this.
The resampling should be a tiny fraction of runtime, so there is no problem doing this say 20-50 times.
Right now, our alpha computation is based on a truth curve created by resampling a much larger NFW truth set to match the redshift distribution of our sample. @mcclearyj has shown this to result in some shot noise in alpha of order a few percent, so we should instead take the mean of multiple subsamples in the
shear-profilesmodule to eliminate this.The resampling should be a tiny fraction of runtime, so there is no problem doing this say 20-50 times.