Skip to content

Benchmarks of how well different two dimensional spaces work for clustering algorithms

Notifications You must be signed in to change notification settings

bramcohen/benchmark_spaces

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Better spatial geometries for use in clustering algorithms

Contains implementations of alternative geometries for spaces which can be used in clustering algorithms. Using the spaces based on optimal sphere packing should result in slightly better affinity heuristics with only minor extra computational requirements. Comparison is here:

Space type comparisons

This metric is a good measure of how much unnecessary noise is added to distances and how easily points can slide past each other when being annealed.

The lattice implementation benchmarks look a little bumpy because of strangeness in sphere packing in different numbers of dimensions. It is likely optimal in dimensions 2 through 8 and close to it in 15 but there may be significant improvements possible in the other numbers of dimensions.

For visualization 2d lattice should be used. It 'crystallizes' nicely as shown here:

Crystallizing in 2d

For applications which need antipodes (opposite points) Torus is best.

About

Benchmarks of how well different two dimensional spaces work for clustering algorithms

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages