number of cluseters and number of classes are orthogonal
oracle may not be classes
look for trim as you increase number of clusters

*  number of times a candidate is choosed as a centroid

*  try to get genic to estimate correct cluseters (EM in weka)
*  results with win-loss-tie
*  lower number of cluseters
*  optimize classification effort (bal)
*  modify genic to output stats on how centroids are moved