User talk:Chase-san/Kd-Tree

NullPointerExceptions and inaccuracy

Trying to test your Kd-Tree in the benchmark framework, I'm getting some NullPointerExceptions due to right.rect.getNearest(k) returning null sometimes. I don't know if it's the proper fix, I changed the relevant lines to be like:

PointKD p = right.rect.getNearest(k);
if(p != null && k.distanceSq(p) < t) {

Something else is wrong also I believe, because it's only getting "88% accuracy" out of the benchmark results, so it's not finding all of the nearest neighbors correctly. --Rednaxela 22:23, 1 March 2010 (UTC)

That should only happen if the rectangle hasn't gotten expanded ever, which shouldn't happen, unless you haven't added anything to the tree at all (darn me and removing all my safeties). Since this gets set during the add. Can you give me some idea of the data you are feeding into it? --Chase 22:44, 1 March 2010 (UTC)

Well, it's adding to the tree before it's searching. See, File:KNN.jar along with Diamond vs CunobelinDC gun data, a recording I made of every kd-tree operation performed by an older version of Voidious's Diamond during a battle. I can upload an updated KNN.jar that includes the testing of your tree if you wish. Personally I've found the framework invaluable for testing my tree, making it easy to see when something I just did added a bug. --Rednaxela 22:55, 1 March 2010 (UTC)

I see, I got it running, and I get the error too. Let me see whats going on. --Chase 23:25, 1 March 2010 (UTC)

Well I found parts of the problem, it gets 100% now, but its about pretty slow (not as slow as Linear search), I suppose that is to be expected, seeing how hard it is on the GC. Still not sure why Rect is returning null, since it shouldn't (probably something stupid per usual). Well I based this new one off my old one, maybe I should write one from scratch. :P --Chase 00:02, 2 March 2010 (UTC)