Here’s another version of my supervised clustering algorithm that saves a few steps using sorting, but it turns out it doesn’t matter much (it seems to be about twice as fast).