Random scattered analysis

copycat/copycat.py

@@ -89,6 +89,7 @@ class Copycat(object):
                 d['avgtemp'] = d.pop('sumtemp') / d['count']
                 d['avgtime'] = d.pop('sumtime') / d['count']
             print('The formula {} provided:'.format(formula))
+            print('Average difference: {}'.format(self.temperature.getAverageDifference()))
             pprint(answers)
 
         return answers

copycat/rule.py

@@ -30,7 +30,7 @@ class Rule(WorkspaceStructure):
             return
         averageDepth = (self.descriptor.conceptualDepth +
                         self.relation.conceptualDepth) / 2.0
-        averageDepth **= 1.1
+        averageDepth **= 1.1 # LSaldyt: This value (1.1) seems 100% contrived.
         # see if the object corresponds to an object
         # if so, see if the descriptor is present (modulo slippages) in the
         # corresponding object
@@ -45,15 +45,15 @@ class Rule(WorkspaceStructure):
                 self.internalStrength = 0.0
                 return
             sharedDescriptorTerm = 100.0
-        conceptual_height = (100.0 - self.descriptor.conceptualDepth) / 10.0
+        conceptual_height = (100.0 - self.descriptor.conceptualDepth) / 10.0 # LSaldyt: 10?
-        sharedDescriptorWeight = conceptual_height ** 1.4
+        sharedDescriptorWeight = conceptual_height ** 1.4 # LSaldyt: 1.4 is also seemingly contrived
         depthDifference = 100.0 - abs(self.descriptor.conceptualDepth -
                                       self.relation.conceptualDepth)
-        weights = ((depthDifference, 12),
+        weights = ((depthDifference, 12),                          # LSaldyt: ???
-                   (averageDepth, 18),
+                   (averageDepth, 18),                             # ????
-                   (sharedDescriptorTerm, sharedDescriptorWeight))
+                   (sharedDescriptorTerm, sharedDescriptorWeight)) # 12 and 18 can be reduced to 2 and 3, depending on sharedDescriptorWeight
         self.internalStrength = formulas.weightedAverage(weights)
-        if self.internalStrength > 100.0:
+        if self.internalStrength > 100.0: # LSaldyt: A better formula wouldn't need to do this.
             self.internalStrength = 100.0
 
     def ruleEqual(self, other):

copycat/temperature.py

@@ -58,7 +58,7 @@ def _averaged_alt(temp, prob):
 
 def _working_best(temp, prob):
     s = .5   # convergence
-    r = 2 # power
+    r = 1.05 # power
     u = prob ** r if prob < .5 else prob ** (1/r)
     return _weighted(temp, prob, s, u)
 
@@ -76,6 +76,8 @@ class Temperature(object):
                 'alt_fifty'      : _alt_fifty,
                 'average_alt'    : _averaged_alt,
                 'best'           : _working_best}
+        self.diffs  = 0
+        self.ndiffs = 0
 
     def reset(self):
         self.actual_value = 100.0
@@ -108,7 +110,14 @@ class Temperature(object):
     def getAdjustedProbability(self, value):
         temp = self.value()
         prob = value
-        return self._adjustmentFormulas[self.adjustmentType](temp, prob)
+        adjusted = self._adjustmentFormulas[self.adjustmentType](temp, prob)
+
+        self.diffs  += abs(adjusted - prob)
+        self.ndiffs += 1
+        return adjusted
+
+    def getAverageDifference(self):
+        return self.diffs / self.ndiffs
 
     def useAdj(self, adj):
         print('Changing to adjustment formula {}'.format(adj))

copycat/workspace.py

@@ -101,8 +101,6 @@ class Workspace(object):
     def getUpdatedTemperature(self):
         '''
         Calculation of global tolerance towards irrelevance
-
-        temp = weightedAverage(totalUnhappiness(.8), ruleWeakness(.2))
         '''
         self.calculateIntraStringUnhappiness()
         self.calculateInterStringUnhappiness()

tests.py

@@ -25,7 +25,6 @@ class TestCopycat(unittest.TestCase):
         self.longMessage = True  # new in Python 2.7
 
     def assertProbabilitiesLookRoughlyLike(self, actual, expected, iterations):
-
         answerKeys = set(list(actual.keys()) + list(expected.keys()))
         degreesFreedom = len(answerKeys)
         chiSquared = 0
@@ -41,7 +40,7 @@ class TestCopycat(unittest.TestCase):
                 chiSquared += (O - E) ** 2 / E
 
         if chiSquared >= _chiSquared_table[degreesFreedom]:
-            self.fail('Significant different between expected and actual answer distributions: \n' +
+            self.fail('Significant difference between expected and actual answer distributions: \n' +
                 'Chi2 value: {} with {} degrees of freedom'.format(chiSquared, degreesFreedom))
 
     def run_testcase(self, initial, modified, target, iterations, expected):