knn: skipped items output when there is a skip function has always the last index and not 0 index

[lukeanto]

Hello, I had old code written for Julia 1.0

using NearestNeighbors
data = [[0.13380863416387367, 0.7845254987714512],[0.1563342025559629, 0.7956456895676077],[0.23320094627474594, 0.9055515160266435]]
tree = KDTree(hcat(map(p -> [p[1], p[2]], data)...))
nearest, distance = knn(tree, [0.15, 0.8], 3, true, x -> x == 2)

In Julia 1.0, nearest is returned as [1, 3, 0], because 2 is skipped; in Julia 1.7.3, nearest is returned as [1, 3, 3], which seems not correct.

Packages status in Julia 1.0

  [15f4f7f2] AutoHashEquals v0.2.0
  [864edb3b] DataStructures v0.18.13
  [c27321d9] Glob v1.3.0
  [cd3eb016] HTTP v0.9.17
  [7073ff75] IJulia v1.23.3
  [6218d12a] ImageMagick v0.7.9
  [02fcd773] ImageTransformations v0.8.13
  [916415d5] Images v0.23.3
  [b8a86587] NearestNeighbors v0.4.10

Packages statuses in Julia 1.7.3:

  [15f4f7f2] AutoHashEquals v0.2.0
  [864edb3b] DataStructures v0.18.13
  [c27321d9] Glob v1.3.0
  [cd3eb016] HTTP v0.9.17
  [6218d12a] ImageMagick v1.2.1
  [02fcd773] ImageTransformations v0.9.4
  [916415d5] Images v0.25.2
  [b8a86587] NearestNeighbors v0.4.10

So they are the same, except for IJulia, which should not be significant anyway

[KristofferC]

It's a bit strange that the Julia version would change this... I'll try to figure out what changed.

Edit: Ah, it is a bounds error...

[KristofferC]

Can you try with #146?

[lukeanto]

Tried just now, this is the final output,

...
julia> nearest, distance = knn(tree, [0.15, 0.8], 3, true, x -> x == 2)
([1, 3], [0.02239688629947563, 0.13440059522389006])

It is different from the original [1, 3, 0], but seems anyway correct to me (probably better, I have not to check for index 0).