User talk:Pedersen/kdTree

I don't have a application for a tree yet (though conceivably it would work with a gun if I bothered to develop weights for the various dimensions, after determining what those dimensions should be) but I figured it would be fun to whip one up. It turns out it was pretty quick to crank out, and after a little debugging I not only had blazing fast speeds at high volumes, faster by far than the samples that came with the benchmark at about 5000 data points, I found that at about that range I started dipping below 100% accuracy. I have spent most of today, longer than it took to write the first draft, trying to find out what is going wrong. As the day winds to a close, I've found that I can get it to bleed accuracy by setting the capacities really low, which will also help with debugging (in terms of diagramming out in code and by hand). When I set the dimensions to 2, neighbors to 10, and sample size to 50, along with a bucket size of 4, I caused it to hit 90% accuracy. My time is up today though, so I'll have to proceed with debugging another evening. --Martin 00:27, 6 March 2010 (UTC)