[Predator] Implement training for loglinear models

appengine/findit/crash/loglinear/test/loglinear_testcase.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 random
+import unittest
+
+from crash.loglinear.feature import FeatureValue
+from crash.loglinear.model import ToFeatureFunction
+
+
+class LoglinearTestCase(unittest.TestCase): # pragma: no cover
+  """Common code for testing ``model.py`` and ``training.py``."""
+
+  def setUp(self):
+    """Set up some basic parts of our loglinear model.
+
+    These parts describe a silly model for detecting whether an integer
+    in [0..9] is the number 7. So ``X`` is the set of integers [0..9],
+    and ``Y`` is the set of ``bool`` values. The independent variable
+    is boolean-valued because we only have two categories: "yes, x ==
+    7" and "no, x != 7". This doesn't take advantage of the fact that
+    loglinear models can categorize larger sets of labels, but it's good
+    enough for testing purposes.
+
+    In addition to specifying ``X`` and ``Y``, we also specify a set of
+    features and choose some random weights for them.
+    """
+    super(LoglinearTestCase, self).setUp()
+
+    # Some arbitrary features.
+    # We don't use double lambdas because gpylint complains about that.
+    def feature0(x):
+      return lambda y: FeatureValue('feature0', y == (x > 5), None, None)
+
+    def feature1(x):
+      return lambda y: FeatureValue('feature1', y == ((x % 2) == 1), None, None)
+
+    def feature2(x):
+      return lambda y: FeatureValue('feature2', y == (x <= 7), None, None)
+
+    self._feature_list = [feature0, feature1, feature2]
+    self._feature_function = ToFeatureFunction(self._feature_list)
+    self._qty_features = len(self._feature_list)
+    self._X = range(10)
+    self._Y = [False, True]
+    self._weights = [random.random() for _ in xrange(self._qty_features)]