[Scipy-svn] r4078 - in trunk/scipy/cluster: . tests

[scipy-svn@scip...]

Author: cdavid
Date: 2008-04-03 13:28:44 -0500 (Thu, 03 Apr 2008)
New Revision: 4078

Modified:
   trunk/scipy/cluster/tests/test_vq.py
   trunk/scipy/cluster/vq.py
Log:
Handle data of rank 1 for random init in kmean2

Modified: trunk/scipy/cluster/tests/test_vq.py
===================================================================
--- trunk/scipy/cluster/tests/test_vq.py	2008-04-03 15:13:35 UTC (rev 4077)
+++ trunk/scipy/cluster/tests/test_vq.py	2008-04-03 18:28:44 UTC (rev 4078)
@@ -131,6 +131,14 @@
         code1 = kmeans2(data1, code, iter = 1)[0]
         code2 = kmeans2(data1, code, iter = 2)[0]
 
+    def test_kmeans2_rank1_2(self):
+        """Testing simple call to kmeans2 with rank 1 data."""
+        data = N.fromfile(open(DATAFILE1), sep = ", ")
+        data = data.reshape((200, 2))
+        data1 = data[:, 0]
+
+        code1 = kmeans2(data1, 2, iter = 1)
+
     def test_kmeans2_init(self):
         """Testing that kmeans2 init methods work."""
         data = N.fromfile(open(DATAFILE1), sep = ", ")

Modified: trunk/scipy/cluster/vq.py
===================================================================
--- trunk/scipy/cluster/vq.py	2008-04-03 15:13:35 UTC (rev 4077)
+++ trunk/scipy/cluster/vq.py	2008-04-03 18:28:44 UTC (rev 4078)
@@ -460,15 +460,28 @@
             Number of samples to generate.
 
     """
-    mu  = N.mean(data, 0)
-    cov = N.atleast_2d(N.cov(data, rowvar = 0))
+    def init_rank1(data):
+        mu  = N.mean(data)
+        cov = N.cov(data)
+        x = N.random.randn(k)
+        x *= N.sqrt(cov)
+        x += mu
+        return x
+    def init_rankn(data):
+        mu  = N.mean(data, 0)
+        cov = N.atleast_2d(N.cov(data, rowvar = 0))
 
-    # k rows, d cols (one row = one obs)
-    # Generate k sample of a random variable ~ Gaussian(mu, cov)
-    x = N.random.randn(k, mu.size)
-    x = N.dot(x, N.linalg.cholesky(cov).T) + mu
+        # k rows, d cols (one row = one obs)
+        # Generate k sample of a random variable ~ Gaussian(mu, cov)
+        x = N.random.randn(k, mu.size)
+        x = N.dot(x, N.linalg.cholesky(cov).T) + mu
+        return x
 
-    return x
+    nd = N.ndim(data)
+    if nd == 1:
+        return init_rank1(data)
+    else:
+        return init_rankn(data)
 
 _valid_init_meth = {'random': _krandinit, 'points': _kpoints}