NearestNeighbors.jl

Data structures for exact and approximate nearest neighbor search. We have:

• A naive search "tree", which uses brute force to find nearest neighbors and caches nothing

The main API for each search structure is:

• Tree construction:
  • t = NaiveNeighborTree(X, Euclidean())
• k nearest-neighbors:
  • nearest(t, v, k)
  • nearest(t, v, k, exclude)
• Neighbors in a ball of radius r:
  • inball(t, v, r)
  • inball(t, v, r, exclude)

Usage Example

using Distance
using NearestNeighbors

X = readcsv(Pkg.dir("NearestNeighbors", "test", "iris.csv"))

t = NaiveNeighborTree(X)

v = X[:, 21]

inds, dists = nearest(t, v, 1)
inds, dists = nearest(t, v, 2)
inds, dists = nearest(t, v, 3)

inds, dists = nearest(t, v, 1, 21)
inds, dists = nearest(t, v, 2, 21)
inds, dists = nearest(t, v, 3, 21)

inds, dists = inball(t, v, 0.5)
inds, dists = inball(t, v, 0.5, 21)

Coming Soon

• Ball-trees
• KD-trees
• Cover-trees
• Approximate nearest neighbor search