Closed jacobgolomb closed 1 year ago
I think the pipelines fail because I changed how pdet is computed so we won't get the same likelihood that the pipeline wants to see.
Also, you should just change the likelihood value in the failing test as we are deliberately changing how the likelihood is calculated.
I would also suggest just always returning
mu
when not applying the correction.
To clarify, you are suggesting we remove the n_eff convergence check here? I'd be wary to do this as I could imagine the sampler oversampling regions of low injection coverage, unless this PR is merged along with the ability to cut on total likelihood uncertainty (which may be the plan, just checking).
This PR changes the default call of the ResamplingVT to use the estimator of the Monte Carlo rather than the marginalized pdet from Farr 2019. Justification: see top of pg 14 on Essick & Farr for example.