[Predator] Fix weight flaky loglinear training test.

appengine/findit/crash/loglinear/test/weight_test.py

Index: appengine/findit/crash/loglinear/test/weight_test.py
diff --git a/appengine/findit/crash/loglinear/test/weight_test.py b/appengine/findit/crash/loglinear/test/weight_test.py
index c1e882395d8f1b90935bb36c3b353c9df2f9e09d..9e3cb0ca7c703f37ae3123e0f8aeb8d02a71ad3a 100644
--- a/appengine/findit/crash/loglinear/test/weight_test.py
+++ b/appengine/findit/crash/loglinear/test/weight_test.py
@@ -60,12 +60,12 @@ class MetaWeightTest(unittest.TestCase):
                      MetaFeatureValue('f', {'f1': 2., 'f2': 1.}), 2.)
     self.assertEqual(MetaFeatureValue('f', {'f1': 0.8, 'f2': 0.4}) *
                      MetaWeight({'f1': Weight(2.), 'f2': Weight(1.)}), 2.)
-    with self.assertRaisesRegexp(Exception,

[stgao, 2017/03/14 04:34:33]

So you mentioned in the bug "when the trained weights have zeros, assertion
exceptions will raise". But in this testcase, do we use randomly-generated
trained weights? If not, I'm wondering how this became flaky? Is it because of
precision issue in floating point wen compared to zero?

[Sharu Jiang, 2017/03/14 05:09:17]

No, the flaky test is ``testTrainWeights`` in training_test.py. It is because
the training data set of loglinear model is randomly generated.

-                                 ('MetaWeight can only multiply with '
-                                  '``MetaFeatureValue`` '
-                                  'with the same length')):
-      dummy_result = MetaWeight({'f1': 0.8, 'f2': 0.4}) * MetaFeatureValue(
-          'f', {'f1': 2.})
+
+    self.assertEqual(
+        MetaWeight({'f1': Weight(0.8), 'f3': Weight(0.0)}) *
+        MetaFeatureValue('f', {'f1': Weight(2.), 'f2': Weight(9),
+                               'f3': Weight(10)}),
+        1.6)
 
   def testIter(self):
     """Tests overloading operator ``__iter__``."""