Implementing loglinear classification (without training), for CL classification

appengine/findit/crash/test/loglinear_test.py

+# Copyright 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import math
+import numpy as np
+import random
+import unittest
+
+from crash.loglinear import ToFeatureFunction
+from crash.loglinear import LogLinearModel
+
+
+# Some arbitrary features.
+# We don't use double lambdas because gpylint complains about that.
+def feature0(x):
+  return lambda y: 1.0 if y == (x > 5) else 0.0
+
+
+def feature1(x):
+  return lambda y: 1.0 if y == ((x % 2) == 1) else 0.0
+
+
+def feature2(x):
+  return lambda y: 1.0 if y == (x <= 7) else 0.0
+
+
+features = [feature0, feature1, feature2]
+X = range(10)
+Y = [False, True]
+
+
+class LoglinearTest(unittest.TestCase):
+
+  def testToFeatureFunction(self):
+    """Test that ``ToFeatureFunction`` obeys the equality its docstring says."""
+    f = ToFeatureFunction(features)
+    for x in X:
+      for y in Y:
+        for i in xrange(len(features)):
+          self.assertEqual(features[i](x)(y), f(x)(y)[i])
+
+  def testLogLinearModel(self):
+    """An arbitrary test to get 100% code coverage.
+
+    Right now this test simply calls every method. The only assertions are
+    that log-domain and normal-domain things are related appropriately;
+    and similarly for the quadrance and l2-norm. Since the one is defined
+    in terms of the other in exactly the way written here, those should
+    be trivially true. However, if the implementation changes, then they
+    may become flaky due to floating point fuzz. Really this should be
+    replaced by a collection of semantically meaningful tests, i.e.,
+    ones that actually look for bugs we might realistically need to
+    guard against. At least this test is good for detecting typo-style
+    errors where we try accessing fields/methods that don't exist.
+    """
+    weights = [random.random() for _ in features]
+
+    model = LogLinearModel(Y, ToFeatureFunction(features), weights, 0.1)
+    model.ClearAllMemos()
+    model = LogLinearModel(Y, ToFeatureFunction(features), weights)
+    self.assertListEqual(weights, model.weights.tolist())
+    self.assertEqual(math.sqrt(model.quadrance), model.l2)
+
+    for x in X:
+      self.assertEqual(math.exp(model.LogZ(x)), model.Z(x))
+      model.Expectation(x, lambda y: np.array([1.0]))
+      for y in Y:
+        model.Features(x)(y)
+        model.Score(x)(y)
+        self.assertEqual(
+            math.exp(model.LogProbability(x)(y)),
+            model.Probability(x)(y))