Making a faster K-NN algorithm

Pretty neat read. Cool to see that something as trivial as the distance measure, can save up on FLOPs so much that it beats scikit's default.