Skip to content

Add bound_snr method to FGMC model fit

Daniel Tang requested to merge feature/bound_snr_in_fit_method into main

This MR adds the bound_snr function to the fit method of the FGMC model TwoComponentModel.

Previously this was only added to the predict method of the FGMC model.

This has the effect on enforcing an upper bound on the input SNR if both the input FAR < FAR_threshold and input SNR > SNR_threshold. Given that this SNR is then passed through to calculate the bayes factor which is in turn used to fit our Poisson posteriors, it may be pertinent to have the same bounded SNR logic during fit as well.

Technically, we should only merge in these changes if they make scientific sense (i.e. consistent data inputs for our statistical model) and improve accuracy.

Alternatively - if reasonable - we might want to only optionally apply bound_snr to fit and not predict, but such an approach does not seem as justified. It is not clear what approach was used by GWCelery during O3 as the method of calculating the "mean counts" dictionary from ligo.p_astro.data is not entirely clear. See GWCelery's p_astro task.

Merge request reports