Skip to main content

Nearest Neighbour Queries

Given a target T and a set of points S, find the nearest neighbour of T in S.

https://www.youtube.com/watch?v=Glp7THUpGow

https://www.youtube.com/watch?v=XG4zpiJAkD4

https://www.cs.cmu.edu/~ckingsf/bioinfo-lectures/kdtrees.pdf

package algorithmsinanutshell.spatialtree

import algorithmdesignmanualbook.print
import kotlin.math.pow


/**
* Given a target T and a set of points S, find the nearest neighbour of T in S.
*
* https://www.youtube.com/watch?v=Glp7THUpGow
*
* https://www.youtube.com/watch?v=XG4zpiJAkD4
*
* https://www.cs.cmu.edu/~ckingsf/bioinfo-lectures/kdtrees.pdf
*/
class NearestNeighbourQueries(val array: Array<Array<Int>>) {

private val tree: MultiDimNode = KDTree(array).tree

private fun findNearest(subTree: MultiDimNode?, target: Array<Int>, depth: Int): MultiDimNode? {
if (subTree == null) return null
val dimensionIndex = depth % target.size

val nextBranch: MultiDimNode?
val otherBranch: MultiDimNode?
if (target[dimensionIndex] < subTree.value[dimensionIndex]) {
nextBranch = subTree.left
otherBranch = subTree.right
} else {
nextBranch = subTree.right
otherBranch = subTree.left
}


var nodeFromNextBranch = findNearest(nextBranch, target, depth + 1)
var best = closest(MultiDimNode(target), nodeFromNextBranch, subTree)

val euclideanDistance = best!!.distanceFrom(target)
val perpendicularDistance = (target[dimensionIndex] - subTree.value[dimensionIndex]).toDouble().pow(2)

// Traverse into the unvisited section i.e otherBranch if the best distance found so far is bigger than
// the perpendicular distance to the unvisited branch
if (euclideanDistance >= perpendicularDistance) {
nodeFromNextBranch = findNearest(otherBranch, target, depth + 1)
best = closest(MultiDimNode(target), nodeFromNextBranch, subTree)
return best
}
return best
}

fun execute(target: Array<Int>): MultiDimNode? {
return findNearest(tree, target, 0)
}

private fun closest(target: MultiDimNode, vararg p1: MultiDimNode?): MultiDimNode? {
return p1.toList()
.filterNotNull()
.minByOrNull {
it.distanceFrom(target.value)
}
}
}

fun main() {
run {
val array = arrayOf(
arrayOf(3, 6), arrayOf(17, 15), arrayOf(13, 15),
arrayOf(6, 12), arrayOf(9, 1), arrayOf(2, 7), arrayOf(10, 19)
)
NearestNeighbourQueries(array).execute(arrayOf(1, 2)).print()
}

run {
val array =
arrayOf(arrayOf(30, 40), arrayOf(5, 25), arrayOf(10, 12), arrayOf(70, 70), arrayOf(50, 30), arrayOf(35, 45))
NearestNeighbourQueries(array).execute(arrayOf(52, 52)).print()
}
}

Updated on 2021-08-22