Skip to content

Latest commit

 

History

History
80 lines (52 loc) · 2.25 KB

README.md

File metadata and controls

80 lines (52 loc) · 2.25 KB
Raw

Lua-kdtree - n-dimensional kdtree spatial indexing.

1. What?

A kdtree is a data structure for quickly finding out what objects in space intersect with a given query volume.

This implements a read-only n-dimensional kdtree that can contain boxes as well as points.

Read-only means that you build the kdtree once on a large set of data, and query many times on that data set. Once the kdtree is built, you can't add objects to it. This could be extended, but it's all I need for the moment.

N-dimensional means that the kdtree works on 2D-spaces (like road maps), a 3D spaces, and higher dimensional spaces. At higher dimensions, spatial querying makes less and less sense though.

The objects in the kdtree can be boxes as well as points. I don't know why, but many of the implementations I've can only contain points.

The kdtree can be written to a file. This can be a text file, or, if you're using LuaJIT, and have ljsycall and lua-mmapfile, a binary file.

2. How?

Either

git clone [email protected]:geoffleyland/lua-kdtree.git
cd lua-kdtree
make install

or, once the rockspec's up luarocks install kdtree. Then:

local kdtree = require"kdtree"

local my_objects = -- make some objects

local function my_bounds(o, min, max)
  min[1], min[2] = o.left, o.bottom
  max[1], max[2] = o.right, o.top
  return min, max
end

local tree = kdtree.build(my_bounds, 2, my_objects)

for o in tree:query({1, 1}, {2, 2}) do
  print("Found", tostring(o))
end

3. Requirements

Lua >= 5.1 or LuaJIT >= 2.0.0, ljsyscall and lua-mapfile for binary file io.

4. Issues

  • The argument list to build is horrible.
  • It would be good to be able to hold object ids (64 bit integers) directly
  • in the tree.

5. Wishlist

  • Read-write kdtrees?
  • A bit of speed work using leaf and node bounds (a few experiments with these didn't make much progress)

6. Alternatives

  • SQLite and PostGres have spatial indexing capabilities. This library running with LuaJIT seems to be about twice as fast at querying as SQLite (on a cheap, nasty, nonrepresentative benchmark)