[Predator] Move ``SingleFeatureScore`` to LLM.

appengine/findit/crash/loglinear/changelist_classifier.py

Index: appengine/findit/crash/loglinear/changelist_classifier.py
diff --git a/appengine/findit/crash/loglinear/changelist_classifier.py b/appengine/findit/crash/loglinear/changelist_classifier.py
index ccbfb569999f418bb0f063a87a73ccc0c6195ef3..3823d9cc147a76bae96a1f474bc3225758158d8e 100644
--- a/appengine/findit/crash/loglinear/changelist_classifier.py
+++ b/appengine/findit/crash/loglinear/changelist_classifier.py
@@ -11,7 +11,7 @@ from crash import changelist_classifier
 from crash.crash_report_with_dependencies import CrashReportWithDependencies
 from crash.loglinear.changelist_features import min_distance
 from crash.loglinear.changelist_features import top_frame_index
-from crash.loglinear.model import ToFeatureFunction
+from crash.loglinear.feature import FeatureFunction
 from crash.loglinear.model import UnnormalizedLogLinearModel
 from crash.stacktrace import CallStack
 from crash.stacktrace import Stacktrace
@@ -42,20 +42,12 @@ class LogLinearChangelistClassifier(object):
     self._top_n_frames = top_n_frames
     self._top_n_suspects = top_n_suspects
 
-    feature_function = ToFeatureFunction([
+    feature_function = FeatureFunction([
         top_frame_index.TopFrameIndexFeature(top_n_frames),
         min_distance.MinDistanceFeature(),
     ])
 
-    weight_list = [
-        weights['TopFrameIndex'],
-        weights['MinDistance'],
-    ]
-
-    self._model = UnnormalizedLogLinearModel(feature_function, weight_list)
-
-    # TODO(crbug.com/674262): remove the need for storing these weights.
-    self._weights = weights
+    self._model = UnnormalizedLogLinearModel(feature_function, weights)
 
   # TODO(crbug.com/673964): something better for detecting "close to log(0)".
   def _LogZeroish(self, x):

[wrengr, 2017/01/11 20:38:30]

This should also be moved to UnnormalizedLLM. That way it can be based on the
same epsilon used there, rather than only checking that it is exactly equal to
-inf.

[Sharu Jiang, 2017/01/12 01:41:38]

Done.

@@ -75,20 +67,6 @@ class LogLinearChangelistClassifier(object):
     """
     return x < 0 and math.isinf(x)
 
-  def _SingleFeatureScore(self, feature_value):
-    """Returns the score (aka weighted value) of a ``FeatureValue``.
-
-    This function assumes the report's stacktrace has already had any necessary
-    preprocessing (like filtering or truncating) applied.
-
-    Args:
-      feature_value (FeatureValue): the feature value to check.
-
-    Returns:
-      The score of the feature value.
-    """
-    return feature_value.value * self._weights.get(feature_value.name, 0.)
-
   def __call__(self, report):
     """Finds changelists suspected of being responsible for the crash report.
 
@@ -142,10 +120,10 @@ class LogLinearChangelistClassifier(object):
 
       suspect.confidence = score
       features = features_given_report(suspect)
-      suspect.reasons = self.FormatReasons(features)
+      suspect.reasons = self.FormatReasons(features.itervalues())
       suspect.changed_files = [
           changed_file.ToDict()
-          for changed_file in self.AggregateChangedFiles(features)]
+          for changed_file in self.AggregateChangedFiles(features.itervalues())]
       scored_suspects.append(suspect)
 
     scored_suspects.sort(key=lambda suspect: suspect.confidence)
@@ -173,7 +151,7 @@ class LogLinearChangelistClassifier(object):
     """
     formatted_reasons = []
     for feature in features:
-      feature_score = self._SingleFeatureScore(feature)
+      feature_score = self._model.SingleFeatureScore(feature)
       if self._LogZeroish(feature_score): # pragma: no cover
         logging.debug('Discarding reasons from feature %s'
             ' because it has zero probability' % feature.name)
@@ -201,7 +179,8 @@ class LogLinearChangelistClassifier(object):
     """
     all_changed_files = {}
     for feature in features:
-      if self._LogZeroish(self._SingleFeatureScore(feature)): # pragma: no cover
+      if self._LogZeroish(
+          self._model.SingleFeatureScore(feature)): # pragma: no cover
         logging.debug('Discarding changed files from feature %s'
             ' because it has zero probability' % feature.name)
         continue