We think that the optimized marginalization schemes will just need the squares of kappa and rho. So we computed these instead of accumulating them straight into the log likelihood, which takes the form
log(L) = const + kappa^2 - (1/2) rho^2
Now we pass kappa_sq and rho_sq into a likelihood helper, which at the moment just returns kappa_sq - 0.5*rho_sq, but we can swap that with (e.g.) the distance-marginalized likelihood.
We think that the optimized marginalization schemes will just need the squares of kappa and rho. So we computed these instead of accumulating them straight into the log likelihood, which takes the form
Now we pass
kappa_sqandrho_sqinto a likelihood helper, which at the moment just returnskappa_sq - 0.5*rho_sq, but we can swap that with (e.g.) the distance-marginalized likelihood.