source: trunk/greenstone3-extensions/vishnu/src/vishnu/cluster/ClusterArray.java@ 8189

package vishnu.cluster;

import java.util.*;


public class ClusterArray
{
    Vector contents;

    ClusterArray(int num_of_clusters)
    {
        int i;

        contents = new Vector(num_of_clusters);

        for (i = 0; i < num_of_clusters; i++)
             contents.addElement(new Cluster(i, null, null, null, 0, 1));
             // i.e. id = i, similarity between kids = 0, #items = 1
    }

    // counts the number of clusters that are less dissimilar than disim
    private void sweep_tree (Cluster cluster, double disim, int min_c_items, TheInt clusters)
    {
        cluster._split = 0;

        if (cluster._id == -1)
        {
            if (cluster._children_sim <= disim && cluster._items >= min_c_items)
            {
                cluster._split = 1;
                contents.addElement(cluster);
                clusters.val++;
            }
            else
            {
                sweep_tree (cluster._child_1, disim, min_c_items, clusters);
                sweep_tree (cluster._child_2, disim, min_c_items, clusters);
            }
        }
        else
        {
            if (cluster._items >= min_c_items)
            {
                cluster._split = 1;
                contents.addElement(cluster);
                clusters.val++;
            }
        }
    }


    double split (Cluster cluster, double[] heights, int n, int target,
                   int min_c_items, TheInt clusters)
    {

  /*
     Try to split 'cluster', made up of 'n' vectors, into 'target'
     clusters by incrementing iteratively the mimimum required
     similarity between the 'cluster\'s' children. The result is an
     array of clusters, 'cluster_ary' which contains 'clusters'
     clusters.  Obviously the final number of clusters may not be the
     same as the desired number of clusters. The function return the
     threshold similarity which resulted in the clusters being split
     in 'clusters' clusters.
  */

      int i;

      if (target < 2)
              target = 2;

      if (target > n)
              target = n;

      // height store minimum similarity values
      // this is then the iterative incrementing loop
      for (i = target - 2; i <= n - 1; i++)
      {
          clusters.val = 0; // simple class for an integer to
                            // be able to pass it by reference (below)
          contents.removeAllElements();
          sweep_tree (cluster, heights[i], min_c_items, clusters);

          if (clusters.val >= target)
                  break;
      }

      // if target - 2 > n - 1, i.e. if the previous loop has not been 
      // entered at all
      if (i > n - 1)
      {
            /* ignore the min_c_size requirement */
            i = target - 1;
            clusters.val = 0;
            contents.removeAllElements();
            sweep_tree (cluster, heights[i], 1, clusters);
      }
      
      return heights[i];
      
    }

}