It's been a crazy few days at work (actually, normal by the crazy standards of my group this time of year, but crazy by any normal standard). Anyway, it's prevented me from acting on the previous post.
And, that's probably a good thing.
I had another meeting with my adviser today and we both agreed that the whole rx4-dimensional space was just a bit nutty. So, we're going to frame this in less grandiose terms. We'll still create a 4-dimensional kernel distribution, but we won't blow it out for every partition. Instead, we'll just sample from that and estimate the block variance. My adviser would like to go even beyond that and just directly sample the total sum variance, but I'm still lobbying for the block variance. I'm pretty sure I can show that it's equivalent, both theoretically and in terms of computational effort. Sticking with block variance gives us a consistent measure of spread throughout the paper.
I've got until Monday to make my case. It's time to wrap this thing up.
No comments:
Post a Comment