Post date: Sep 26, 2014 2:13:13 PM
I've spent a fair amount of time this week trying to speed up Bungee by adding some parallel processing to the mix. Today I got to work and found out that it seems to be working, and with a speed up in one section of the code of about twice using two additional cores. It's not great, but when the step I sped up takes sometimes hours to complete, even a factor of two is respectable. I plan to parallelize some other code in the coming weeks.
At the same time, I've continued to work on my third chapter. This week I actually sat down and started writing the damn thing... yes, on paper with pen. I've found it terribly distracting to try to do this particular paper on the computer, and I've had somewhat more success just putting pen to paper and working through it.
Coming out of my meeting with my adviser last week, I decided I needed to look at the spatial scales over which costs and impacts from Bungee are varying. To that end, I learned about this super cool tool called variograms, which show how things vary in space as you look over larger and larger scales. By creating a variogram for costs and another for impacts, I can determine what scales are most important to each, and consequently how the degree of trandeoffs of impacts and costs depends on the spatial scale of their variances. Here's a quick Excel variogram from one of my third chapter case study sites:
Hmmm, well now that doesnt look very interesting. The two things are almost identical, which means impacts and costs vary almost exactly the same over any scale... now it could be that if I continue this thing out past 600 meters I'll see something a bit more interesting. The expectation for the results I am finding (not shown) is that costs vary less than impacts over small scales. But out at 600 meters is pretty far, so I still need to do a bit more digging to find out what's going on.