changing the init method to Furthest. PlusPlus seemed to still be una…

…ble to find the global minima, even with 4 trials. (with this dataset). Furthest hasn't failed for me yet (I pick best of 4 trials). I wonder if PlusPlus is a biased init, so it's not as interesting if you need to pick best of N trials to get best result.

py/testdir_multi_jvm/test_KMeans2_sphere5_bad_inits.py

@@ -115,7 +115,8 @@ def test_KMeans2_sphere5_bad_inits(self):
                 'normalize': 0,
                 'k': CLUSTERS, 
                 'max_iter': MAX_ITER, 
-                'initialization': 'PlusPlus',
+                'initialization': 'Furthest',
+                # 'initialization': 'PlusPlus',
                 'destination_key': 'syn_spheres100.hex', 
                 'seed': SEED
             }