这个算法的基础是误差平方和准则，为了要得到最优结果，首先要对样本集进行划分，一般的做法是先选择一些代表性的点作为聚类的核心，然后把其余的点按照某种方法分到各类中去。-This algorithm is based on error sum of squares criterion, in order to get the optimal results, first of all to sample set division, the general practice is to choose a few representative point as the core of the clustering, and then put the rest of the point according to some means to all kinds of to.