@rjnw how fast does it sample by comparison to us?

I don’t know what counts as a sweep for augur. The way I used, it did 100 sweeps in around 1 second where as our 100 sweeps is 30 seconds. (Every other line on the graph is 100 sweeps from what I remember.)

I’m confused — isn’t it basically running the same model as us?

Rajan and I talked a bit. We think these results confirm that AugurV2 is basically a very fast JAGS, performing 100 sweeps in a couple of seconds. The math involved in AugurV2 sweeps is same as the math involved in JAGS sweeps, but different from the math in our sweeps because we integrate out latent variables. So it’s not particularly bad that we spend more time per our-sweep than AugurV2 spends per AugurV2-sweep.

So we can move on to applying AugurV2 to Naive Bayes for now.

(Oh but Rajan if it’s easy to measure AugurV2 startup times then it would make sense to ad them to Table 1)