for a new point, find the k closest points, make them vote to classify by their labelefficiency bad for compute allWorks well provided there is lots of data and the distance function is good