Reading time ~4 minutes This post is an analysis of a very interesting optimization proposed by Nicholas Frechette in the comments under the previous post. He proposed to use one of the oldest tricks in performance cookbook - divide and conquer. Well, it did not turn out as I expected.Saga Before I go further here are some link to the previous posts on the problem of calculating similarities and then optimizing. This thread grew to a few post. Here are all of them:How I calculate similariti...