[Predator] Add MetaWeight and MetaFeatureValue to group multiple weights and features together.

appengine/findit/crash/loglinear/weight.py

+# Copyright 2017 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
+
+from crash.loglinear.feature import MetaFeatureValue
+from libs.meta_object import Element
+from libs.meta_object import MetaDict
+
+
+class Weight(Element):
+  """Float-like class that represents the weight for a feature."""
+
+  def __init__(self, value):
+    super(Weight, self).__init__()
+    self._value = float(value)
+
+  @property
+  def value(self):
+    return self._value
+
+  def __mul__(self, number):
+    return Weight(self._value * float(number))
+
+  __rmul__ = __mul__
+
+  def __len__(self):
+    return 1
+
+  def __float__(self):
+    return self.value
+
+  def __eq__(self, other):
+    return self._value == other._value
+
+  def __ne__(self, other):
+    return not self.__eq__(other)
+
+  @property
+  def l0(self):
+    """The l0-norm of the weight.
+
+    N.B., despite being popularly called the "l0-norm", this isn't
+    actually a norm in the mathematical sense."""
+    return float(bool(self._value))
+
+  @property
+  def l1(self):
+    """The l1 (aka: Manhattan) norm of the weight."""
+    return math.fabs(self._value)
+
+  @property
+  def quadrance(self):
+    """The square of the l2 norm of the weight.
+
+    This value is often more helpful to have direct access to, as it
+    avoids the need for non-rational functions (e.g., sqrt) and shows up
+    as its own quantity in many places. Also, computing it directly avoids
+    the error introduced by squaring the square-root of an IEEE-754 float.
+    """
+    return math.fabs(self._value)**2
+
+  def IsZero(self, epsilon):
+    return math.fabs(self._value) <= epsilon
+
+
+class MetaWeight(MetaDict):
+  """Dict-like class mapping features in ``Metafeature`` to their weights."""
+
+  def IsZero(self, epsilon):
+    """A MetaWeight is zero only when all the sub weights are zeros."""
+    return all(weight.IsZero(epsilon) for weight in self.itervalues())
+
+  def DropZeroWeights(self, epsilon=0.):
+    """Drops all zero weights."""
+    self._value = {name: weight for name, weight in self.iteritems()
+                  if not weight.IsZero(epsilon)}
+
+  def __len__(self):
+    return len(self._value)
+
+  def __mul__(self, meta_feature):
+    """``MetaWeight`` can multiply with ``MetaFeatureValue``."""
+    assert len(self) == len(meta_feature), Exception(
+        'MetaWeight can only multiply with ``MetaFeatureValue`` with the '
+        'same length')
+
+    # MetaWeight is a dense representation of a sparse array.
+    return math.fsum(meta_feature[name] * weight
+                     for name, weight in self.iteritems())
+
+  __rmul__ = __mul__
+
+  def __eq__(self, other):
+    if len(self) != len(other):
+      return False
+
+    for key, value in self.iteritems():
+      if value != other.get(key):
+        return False
+
+    return True
+
+  def __ne__(self, other):
+    return not self.__eq__(other)
+
+  @property
+  def l0(self):
+    """The l0-norm of the meta weight.
+
+    N.B., despite being popularly called the "l0-norm", this isn't
+    actually a norm in the mathematical sense."""
+    return math.fsum(weight.l0 for weight in self.itervalues())
+
+  @property
+  def l1(self):
+    """The l1 (aka: Manhattan) norm of the meta weight."""
+    return math.fsum(weight.l1 for weight in self.itervalues())
+
+  @property
+  def quadrance(self):
+    """The square of the l2 norm of the meta weight.
+
+    This value is often more helpful to have direct access to, as it
+    avoids the need for non-rational functions (e.g., sqrt) and shows up
+    as its own quantity in many places. Also, computing it directly avoids
+    the error introduced by squaring the square-root of an IEEE-754 float.
+    """
+    return math.fsum(weight.quadrance for weight in self.itervalues())