CGAL users discussion list

["Sebastien Loriot (GeometryFactory)" <>]

Could you provide a minimal example showing the issue?

Sebastien.

On 12/01/2013 11:29 PM, Samaras wrote:
> To make matters worse, here is another behavior with the same distance class
> as mentioned above, that I don't understand.
>
> Now I have a dataset in 22 dimensions with 110K number of points. Then I
> produce 1000 random query points in the cube [-1000, 1000]^D (have tried
> with 1 and 2, instead of 1000, but the behavior seems to be the same).
>
> If I use the splitting rule Midpoint_of_max_spread for the kd_tree, the tree
> is smaller. The strange fact is that now the time measurements seem to be
> pretty much the same, regardless epsilon. With the default rule (i.e.
> Sliding_midpoint), we have (roughly) the behavior mentioned in the previous
> post.
>
> default
> Number of nodes: 69441
>   Tree depth: 53
> NN        0.061071443        0.065089808        0.065244869
> 0.066301312
> FN        0.057983228        0.051108550        0.047526043
> 0.042351644
>
> Midpoint_of_max_spread
> Number of nodes: 59123
>   Tree depth: 37
> NN        0.057491852        0.058924140        0.059315090
> 0.058354231
> FN        0.058201980        0.056180041        0.056738955
> 0.058448647
>
> Two questions, that I can not answer myself and I am afraid no one from here
> is able to do it too! :/
>
>
>
> --
> View this message in context: 
> http://cgal-discuss.949826.n4.nabble.com/Nearest-neighbor-query-time-increases-as-epsilon-increases-tp4658494p4658498.html
> Sent from the cgal-discuss mailing list archive at Nabble.com.