I enjoyed a quirk in this paper by Ngo et al. (2015) on today’s astro ph: the authors use an affine invariant MCMC sampler to estimate a single parameter (the companion fraction) from a single parameter model for which (see plot below) importance sampling from their uniform prior would have had an efficiency of ~50% in terms of effective sample size—and I imagine an ~500% efficiency improvement in terms of human effort (i.e., one line in R vs. the time wasted configuring the MCMC sampler and checking convergence)!

Still, I guess it gives a citation to the authors of the sampler package …

### Like this:

Like Loading...

*Related*

Ok, sure – but a fine sledgehammer it is! 🙂 (Presumably, good exercise for a student)

This post needed some good hammer based puns. A missed opportunity 😉