Post date: Oct 3, 2014 1:54:54 PM
I've finally gotten parallel processing of additive impact surfaces working in Bungee after a couple of weeks working on it. I've been struggling with figuring out how to do useful parallel processing for the creation of layouts. The "embarassingly" parallel version would be to have each processor create some subset of layouts. However, the optimization method relies on iterative improvements from layout to layout, which means every layout (potentially) affects the next one. Now that I say that, though, it's not guaranteed that every layout will be an improvement over the last... so maybe I can have processors report in when they get an improvement. hmmm...
I've also been doing a bit more with variograms. The patterns seem to be more or less the same no matter how I cut it, so that's a pretty strong indication to me that it's not the scale of variation in impacts/costs that's important. Now that I say that, I wonder if the shape of the variogram curve is the whole story. What I might need to show is that the variation within each of cost and impact is different (specifically that impact is more variable than cost at a particular scale). hmmmm...