【摘要】 K-means聚类算法会收敛到求解问题的多个局部最优解中的一个,而且它对初始条件十分敏感。给出了一种基于数据集的多个子集和山函数的初始条件方法,它可以较稳定地收敛到一个更好的局部最优解。此初始方法同时适用于原空间和核空间的K-means算法,相对作用于完整数据集的山函数方法,该方法的时间复杂度和空间复杂度都只有它的p2(p是采样率,p<1)。更多还原

【Abstract】 K-means clusering algorithm converges to one of numerous best local solutions and it is especially sensitive to initial conditions.The refining mountain method is presented for numbers of subsets and Mountain functions based on the data set,and the method can stably converge to a better local solution.The initial method applies to the original space and kernel space K-means algorithms,and it relatively functions on the full data Mountain functions method.The refining mountain method time complexity and space complexity are based on its p2.(p is sample rate,p<1)更多还原