My R-Tree is fast enough for me:) There is on robowiki rough comparsion of range search vs kNN search.
Skilgannon for sure can create faster R-Tree, i think:)
I don't think, that there is fair way to compare speed of R-Tree and KD-Tree, because quite different purpose of them - R-Tree optimised to search for unknown amount of points in strict region, while KD-Tree optimised to search for exact amount of points in unbounded region.
What do CHILDREN_COUNT, and RTreeEntry, each do? Can they be optimized like bucket size? Do you see any optimizations to the R-tree Jdev?
CHILDREN_COUNT - it's split factor, or how many subrectangles will be generated, when current rectangle are overflowed
RTreeEntry... hmmmm... it's entry of range search tree:) It represents data point in turn snapshots space and thing that are added to tree and search for which tree are used.
I think, that CHILDREN_COUNT can be optimised for particular spaces based on theirs natures
I don't know. It was 4 years ago and i do not remember anything
Here's some words to use ctrl f on to help you remember
Rtree CHILDREN_COUNT RTree parent Attribute dimensions IntervalDouble coveredRange RTreeEntry RTreeEntry entries nextEntryIdx RTree children nextChild Intersection intersection splitDimensionIdx singular entryCount TurnSnapshot location