0

I want to implement a Voronoi k-nearest neighbor solution from "Nearest neighbors in high-dimensional data?" I am currently using MATLAB's voronoin to calculate Voronoi vertices and bins. This function uses Qhull.

The problem is that with over 5 million states across 6 dimensions, the function just hangs. Is there a way to limit Qhull to n numbre of vertices? If not, is there an alternative path I can take?

Community
  • 1
  • 1
Elpezmuerto
  • 5,391
  • 20
  • 65
  • 79
  • Are you interesting only in Matlab solutions? – gsamaras Jun 21 '14 at 23:30
  • @G.Samaras I am interested in a solution that can at least interface with MATLAB. Feel free to post any solution though, I can always (maybe) adapt it to MATLAB. – Elpezmuerto Jun 23 '14 at 18:30
  • One solution that comes to mind is not using Voronoi, but a tree-based approach that will find nearest neighbours. Are you interested? – gsamaras Jun 23 '14 at 18:45
  • Reading the algorithm, it seems that you don't really need to compute the polyhedra of the Voronoi Tessalation... To know in which polyhedron a point is, you just need to compute the 250 distances to the germs, and pick the nearest one. – francis Jul 04 '14 at 21:33

0 Answers0