Skip to content

corrected missing sigma factor in posterior sampler

VINAYA VALSAN requested to merge vinaya.valsan/rapidpe-rift-pipe:sampling_correction into main
  1. Rapidpe samples are generated with Gaussians of standard deviation=grid_size and amplitude=Marginlaized_Likelihood. During the implementation, we call numpy.random.normal function which has an amplitude 1/\sqrt{2\pi\sigma} due to normalization. So, wherever we modify the script to have the amplitude=Marginlaized_Likelihood, need a correction factor of \sqrt{2\pi\sigma}.
  1. Rapidpe posterior samples are generated in mchirp-q (or mchirp-mtotal) with a prior uniform in mass1-mass2 as a function of mchirp-q(or mchirp-mtotal). To get the mass1, and mass2 samples we need to convert the mchirp-q (or mchirp-mtotal) sample to mass1-mass2 and assign the prior of the corresponding mchirp-q(or mchirp-mtotal) as a weight for each samples.
  • corrected the weight of m1-m2 samples in the posterior plot
  1. We usually save weighted samples. Additionally save posterior samples after resampling from weighted samples.
  • save posterior samples in addition to weighted posterior samples

Review wiki with these corrections applied: https://git.ligo.org/rapidpe-rift/rapidpe_rift_review_o4/-/wikis/AMR-Review/Review-Version-2

Review wiki before these corrections: https://git.ligo.org/rapidpe-rift/rapidpe_rift_review_o4/-/wikis/AMR-Review

Edited by VINAYA VALSAN

Merge request reports