* Add a basic test suite.

* Fix empty clusters before you add any items
* Fix mass=1, start with mass=0.

Fixes #1

.gitignore

@@ -1,3 +1,6 @@
 *.pyc
 build/
 pygvm.egg-info
+.tox
+dist
+MANIFEST

pygvm/__init__.py

@@ -9,7 +9,7 @@
 import sys
 from array import array
 
-VERSION = (0, 3, 'x')
+VERSION = (0, 3, 'a')
 
 
 MAX_FLOAT = sys.float_info.max
@@ -18,7 +18,7 @@
 class Cluster(object):
     def __init__(self, clusters):
         self.removed = False
-        self.mass = 1.0
+        self.mass = 0.0
 
         # mass-weighted coordinate sum
         self.m1 = [0] * clusters.dimension
@@ -32,9 +32,9 @@ def __init__(self, clusters):
     @property
     def center(self):
         if self.mass:
-            return [coord / self.mass for coord in self.m1]
+            return tuple([coord / self.mass for coord in self.m1])
         else:
-            return self.m1[:]
+            return None
 
     def __len__(self):
         return len(self.members)
@@ -274,9 +274,6 @@ def __init__(self, dimension, capacity, cluster_factory=Cluster):
         self.clusters = []
         self.pairs = ClusterPairs()
         self.cluster_factory = cluster_factory
-        for i in range(capacity):
-            self.clusters.append(self.cluster_factory(self))
-            self._add_pairs()
 
     def clear(self):
         self.clusters = []
@@ -292,6 +289,17 @@ def add_bulk(self, items, step=1000):
 
         This method can be much faster than calling add() repeatedly.
         """
+        allow_merge = True
+        if len(self.clusters) < self.capacity:
+            # create some clusters
+            num_to_add = min(len(items), self.capacity - len(self.clusters))
+            for i in range(num_to_add):
+                self.clusters.append(self.cluster_factory(self))
+
+            allow_merge = len(self.clusters) == self.capacity
+            if allow_merge:
+                # finished adding clusters, so we can populate pairs.
+                self._add_pairs()
 
         # doing individual add/append for each key added is slow,
         # so we delay it until a bunch of things have been added instead.
@@ -301,6 +309,7 @@ def _add_members():
             for c, new_members in cluster_keys.items():
                 c.add_members(new_members)
                 del cluster_keys[c]
+
         try:
             for i, (mass, coords, key) in enumerate(items):
                 if mass == 0:
@@ -313,16 +322,22 @@ def _add_members():
                         break
                 else:
                     c = None
-                if c:
+
+                if c is not None:
+                    # a cluster is empty. add to that cluster and continue
                     c.add(mass, coords, None)
                     if key is not None:
                         cluster_keys[c].append(key)
                     self._update_pairs(c)
                     continue
 
-                #identify cheapest merge
-                merge_pair = self.pairs.peek()
-                merge_t = merge_pair and merge_pair.value or MAX_FLOAT
+                if allow_merge:
+                    # identify cheapest merge
+                    merge_pair = self.pairs.peek()
+                    merge_t = merge_pair.value if merge_pair else MAX_FLOAT
+                else:
+                    # never merge if we still haven't assigned at least one point per cluster.
+                    merge_t = MAX_FLOAT
 
                 # find cheapest addition
                 addition_c = None

pygvm/tests.py

@@ -0,0 +1,40 @@
+import unittest2
+from pygvm import Clusters
+
+class TestClusters(unittest2.TestCase):
+    def test_empty_clusters(self):
+        cs = Clusters(dimension=2, capacity=2)
+
+        self.assertEqual(len(cs.clusters), 0)
+
+    def test_first_point(self):
+        cs = Clusters(dimension=2, capacity=2)
+
+        cs.add(1, (5, 5), 'bert')
+        self.assertEqual(len(cs.clusters), 1)
+        self.assertEqual(len(cs.clusters[0]), 1)
+
+        self.assertEqual(cs.clusters[0].mass, 1)
+        self.assertEqual(cs.clusters[0].center, (5, 5))
+
+    def test_same_point_multiple_times(self):
+        # NOTE: this is stupid. If you seed both clusters with the same
+        # point, your clustering will basically just be random.
+        cs = Clusters(dimension=2, capacity=2)
+
+        cs.add(1, (5, 5), 'tom')
+        self.assertEqual(len(cs.clusters), 1)
+        self.assertEqual(len(cs.clusters[0]), 1)
+
+        cs.add(1, (5, 5), 'dick')
+        self.assertEqual(len(cs.clusters), 2)
+        self.assertEqual(len(cs.clusters[0]), 1)
+        self.assertEqual(len(cs.clusters[1]), 1)
+
+        cs.add(1, (5, 5), 'harry')
+        self.assertEqual(len(cs.clusters), 2)
+        self.assertEqual(len(cs.clusters[0]), 2)
+        self.assertEqual(len(cs.clusters[1]), 1)
+
+        self.assertEqual(cs.clusters[0].mass, 2)
+        self.assertEqual(cs.clusters[1].mass, 1)

setup.py

@@ -11,7 +11,7 @@
     author='Craig de Stigter',
     author_email='craig.ds@gmail.com',
     url='http://github.com/craigds/pygvm',
-    packages=['pygvm', 'pygvm.libs'],
+    packages=['pygvm'],
     classifiers = [
         "Development Status :: 4 - Beta",
         "Environment :: Web Environment",

tox.ini

@@ -0,0 +1,24 @@
+[tox]
+envlist=
+  py26,
+  py27,
+  py33,
+  py34
+
+
+[testenv]
+deps =
+  unittest2
+commands = unit2.py pygvm.tests {posargs}
+
+[testenv:py26]
+basepython=python2.6
+
+[testenv:py27]
+basepython=python2.7
+
+[testenv:py33]
+basepython=python3.3
+
+[testenv:py34]
+basepython=python3.4