2 comments

[ 3.1 ms ] story [ 17.4 ms ] thread
I wonder how it performs compared to a VP tree. Yes, the VP tree has all the problems of the KD tree, i.e. it becomes unbalanced when inserting, but most fast implementations use linear search in leaf nodes instead of creating a full tree, and splitting a leaf once it exceeds its max size (which is normally around 100-1000) usually gives a pretty good vantage point. I guess the same can be done on a KD tree. In my experiments, the VP tree was the fastest tree for range queries when you can't establish a total order over your data.