BMH preliminary results

Here's a first look at the Bootstrap Metropolis-Hastings results. Vertical scales are the same as before.

Cauchy is nutty as usual but the other two are actually pretty good. Then we try a filtered query...

Oh, wowsers, that's not nearly so good. The confidence interval misses the true value most of the time, even in the normal case. The problem, of course, is the correlation of the attributes. Here's the kicker though. Metropolis-Hastings can handle arbitrarily complex models. Currently, I've got it assuming iid inputs. That's not necessary. A model that assumes correlation in the data set might do much better. I'll work on that next.