Builder Workflow

Every tree is built with buildtree (or buildforest), configured by a builder object that carries all construction settings. The builder's type selects the tree type; changing tree type means passing a different builder, not calling a different function.

buildtree(points; builder=TwoNTreeBuilder())
buildtree(testpoints, trialpoints; builder=BlockTreeBuilder())   # -> BlockTree
buildforest(points, graph, weights; builder=MetisForestBuilder())

Direct constructors like TwoNTree(points; builder=...) are convenience wrappers around the same builder; prefer buildtree/buildforest when the tree type should follow from the builder alone.

Builders

The high-level builders have sane defaults: TwoNTreeBuilder(), KMeansTreeBuilder(), MetisTreeBuilder(), and MetisForestBuilder() all work with no arguments. The generic BoundingBallTreeBuilder is lower-level: pass an explicit splitter and split count, or use KMeansTreeBuilder/MetisTreeBuilder for the packaged ball-tree strategies. See Tree Families for which one to pick.

Example: same points, different builder

using H2Trees

m = meshsphere(1.0, 0.1)
points = vertices(m)

adaptive = buildtree(points; builder=TwoNTreeBuilder(; minvalues=60))
uniform = buildtree(points; builder=TwoNTreeBuilder(; minhalfsize=0.1, minvalues=0))
println(adaptive)
println(uniform)

TwoNTree{3,Float64}(nodes=66, leaves=57, levels=1:3, root=1)
TwoNTree{3,Float64}(nodes=510, leaves=360, levels=1:5, root=1)

Swapping in a clustering builder (with ParallelKMeans.jl loaded) needs no other code change:

using ParallelKMeans
clustered = buildtree(points; builder=KMeansTreeBuilder(; numberofclusters=8))

An unsupported (input, builder) combination throws an ArgumentError instead of silently picking a default.