seckcoder
diff --git a/‎scikits/learn/cluster/k_means_.py‎
Lines changed: 19 additions & 7 deletions b/‎scikits/learn/cluster/k_means_.py‎
Lines changed: 19 additions & 7 deletions
diff --git a/‎scikits/learn/metrics/pairwise.py‎
Lines changed: 4 additions & 5 deletions b/‎scikits/learn/metrics/pairwise.py‎
Lines changed: 4 additions & 5 deletions
@@ -184,9 +184,13 @@ def k_means(X, k, init='k-means++', n_init=10, max_iter=300, verbose=0,
         if verbose:
             print 'Initialization complete'
         # iterations
+        x_squared_norms = X.copy()
+        x_squared_norms **=2
+        x_squared_norms = x_squared_norms.sum(axis=1)
         for i in range(max_iter):
             centers_old = centers.copy()
-            labels, inertia = _e_step(X, centers)
+            labels, inertia = _e_step(X, centers,
+                                        x_squared_norms=x_squared_norms)
             centers = _m_step(X, labels, k)
             if verbose:
                 print 'Iteration %i, inertia %s' % (i, inertia)
@@ -228,12 +232,18 @@ def _m_step(x, z, k):
         The resulting centers
     """
     dim = x.shape[1]
-    centers = np.repeat(np.reshape(x.mean(0), (1, dim)), k, 0)
+    centers = np.empty((k, dim))
+    X_center = None
     for q in range(k):
-        if np.sum(z == q) == 0:
-            pass
+        this_center_mask = (z == q)
+        if not np.any(this_center_mask):
+            # The centroid of empty clusters is set to the center of
+            # everything
+            if X_center is None:
+                X_center = x.mean(axis=0)
+            centers[q] = X_center
         else:
-            centers[q] = np.mean(x[z == q], axis=0)
+            centers[q] = np.mean(x[this_center_mask], axis=0)
     return centers
 
 
@@ -265,8 +275,10 @@ def _e_step(x, centers, precompute_distances=True, x_squared_norms=None):
     if precompute_distances:
         distances = euclidean_distances(centers, x, x_squared_norms,
                                         squared=True)
-    z = -np.ones(n_samples).astype(np.int)
-    mindist = np.infty * np.ones(n_samples)
+    z = np.empty(n_samples, dtype=np.int)
+    z.fill(-1)
+    mindist = np.empty(n_samples)
+    mindist.fill(np.infty)
     for q in range(k):
         if precompute_distances:
             dist = distances[q]
 
@@ -8,10 +8,7 @@
 
 import numpy as np
 
-
-def euclidean_distances(X, Y,
-        Y_norm_squared=None,
-        squared=False):
+def euclidean_distances(X, Y, Y_norm_squared=None, squared=False):
     """
     Considering the rows of X (and Y=X) as vectors, compute the
     distance matrix between each pair of vectors.
@@ -59,7 +56,9 @@ def euclidean_distances(X, Y,
     if X is Y: # shortcut in the common case euclidean_distances(X, X)
         YY = XX.T
     elif Y_norm_squared is None:
-        YY = np.sum(Y * Y, axis=1)[np.newaxis, :]
+        YY = Y.copy()
+        YY **= 2
+        YY = np.sum(YY, axis=1)[np.newaxis, :]
     else:
         YY = np.asanyarray(Y_norm_squared)
         if YY.shape != (Y.shape[0],):