So these graphs don’t seem to be so great? Well, I guess for GmmGibbs[9,12,15]–1000, it seems to show that rkt "gets there"in about .2 seconds, quite a bit faster than the others. But I don’t understand why these graphs are the ‘right’ ones to show! Also, why are the lines of so very different jaggedness? It sorts of feels like quite different things might be being measured.

For example, I would have thought to measure: 1) time / sweep for each system, and 2) time to first sweep (warm-up) and 3) time to reach within 5% on long-term medium accuracy.

different jaggedness is due to more number of sweeps within that time period. Here rkt does a lot of sweeps in same time when compared to haskell and jags.

I don’t know if these graphs are the right ones to show, I am just trying to have some visual data so we can figure out what to show.

500 trials with each snapshot in every trial plotted as a single dot with alpha 0.1