03:43 < naywhayare> george___: you're right, that distance is a metric; I was thinking about it wrong

03:43 < naywhayare> can you tell me more about the issue you were having? you could report it as a bug on Trac and I could spend some more time digging into what's wrong, if you like

03:52 < george___> sure, apologies in advance if I describe things in a difficult to understand way

03:53 < george___> I am using knn search to solve an n-body problem with a topological interaction, meaning that my particles are to interact with only their K nearest neighbors

03:54 < george___> in order to use mlpack, I wrote an instance of the metric class, which calculates the distance on the torus

03:55 < george___> everything seemed to work correctly, except when clumps of particles reached the boundary of the torus, where it became clear that particles would not interact with other particles if they were across the edge of the torus

03:56 < george___> I think this is because of the binary space partitioning, but I haven't thought enough about the actual KNN search algorithms to know this is true

03:57 < george___> for instance, if you imagine splitting space in half, many particles might get stuck on the wrong side of the tree

03:58 < george___> imagining the torus (I might be wrong about this)

03:58 < george___> I can make and send a short program that demonstrates this (i.e. my metric class and a computation of neighbors)

03:59 < george___> incidentally, if I set the number of neighbors to find to the maximum (i.e. I have 200 particles and I find the 199 nearest neighbors)

03:59 < george___> this works and I can take the first 25 of these

04:00 < george___> but if I search for the 25 closest, I don't get the correct answer

04:00 < george___> for particles near an edge

04:01 < naywhayare> george___: yeah, if you can send a short demonstration program I can dig to the bottom of what's wrong pretty easily

04:01 < george___> incidentally, I tried to use cover tree, but could not quite get it to work, I think because I don't understand C++/mlpack well enough

04:01 < george___> ok, I'll make that, how should I send it to you?

04:01 < naywhayare> can you file a bug on Trac? you can attach it there; http://trac.research.cc.gatech.edu/fastlab/

04:02 < george___> sure, I'll do that

04:02 < naywhayare> I won't be able to play with it until Monday at the earliest; I'm out of town right now, until tomorrow night

04:03 < naywhayare> thanks :) when I wrote the allknn implementation some time back I took care to write it in a flexible way to allow other metrics, so I am interested in solving any bugs people encounter :)

04:03 < george___> yeah, thanks for looking into this!

04:05 < naywhayare> sure, no problem

04:05 < naywhayare> you can hang out in IRC and I'll update you on what I find, or I'll do it through trac

06:08 < george___> I submitted a ticket, thanks for any help!

06:10 < george___> If you are able to easily modify my code so that it uses CoverTree rather than BinarySpaceTree, let me know how you did it (I think I was insufficiently knowledgable to easily figure this out, as my efforts to use CoverTree were not successful)