A java implementation of k-means algorithm.It uses ball tree as internal data structure to accelerate the computation.It uses 2-norm distance to compute the similarity between instances.

1. The input dataset should have certain form:one line one instance,and neighboured feature should be splitted by space.

2. If you can change the maximum number of instances in a leaf node in the ball tree or the times of running the algorithm to find a better cluster for one specific k,change the conf.properties file in KMeansCluster.

3. The function you can use:
   In the class Process:

   //gives your dataset's path and this function will build the internal data structures. public static void loadData(String path) throws IOException;

   /** * @param k the initial number of clustering centers

   • @return an entry:the key is the result of clustering.The label starts from 0.The value is the evaluation of the clustering result */ public static Entry<Integer[], Double> cluster(int k);

   /**gives the evaluation and differential of each k in specific range.you can use these infos to choose a good k for your clustering

   • @param startK gives the start point of k for the our try on k(inclusive)
   • @param endK gives the end point(exclusive)
   • @return Entry's key is the evaluation of clustering of each k.The value is the differential of the evaluations--evaluation of k(i) - evaluation of k(i+1) for i in range(startK, endK - 1)
   • */ public static Entry<ArrayList, ArrayList> cluster(int startK, int endK);

Before call the last two functions,the first one must be called.

This repository is migrated from my older Github account @skyline0623: link