Probability distributions metrics

benchmarks/ProbabilityDistribution.py

@@ -43,6 +43,16 @@ class ProbabilityDistribution(object):
                   for k in self.bins])
         return h
 
+    def crossentropy(self,q):
+        h = -sum([self.distribution[k] * np.log(q.distribution[k]) if self.distribution[k] > 0 else 0
+                  for k in self.bins])
+        return h
+
+    def kullbackleiblerdivergence(self,q):
+        h = sum([self.distribution[k] * np.log(self.distribution[k]/q.distribution[k]) if self.distribution[k] > 0 else 0
+                  for k in self.bins])
+        return h
+
     def empiricalloglikelihood(self):
         _s = 0
         for k in self.bins:
@@ -60,6 +70,16 @@ class ProbabilityDistribution(object):
                 _s += np.log(k)
         return _s
 
+    def averageloglikelihood(self, data):
+
+        densities = self.density(data)
+
+        _s = 0
+        for k in densities:
+            if k > 0:
+                _s += np.log(k)
+        return _s / len(data)
+
     def plot(self,axis=None,color="black",tam=[10, 6]):
         if axis is None:
             fig = plt.figure(figsize=tam)

benchmarks/benchmarks.py

@@ -119,7 +119,7 @@ def getPointStatistics(data, models, externalmodels = None, externalforecasts =
             ret += " 1		& "
             ret += str(round(Measures.rmse(data, externalforecasts[k][:-1]), 2)) + "		& "
             ret += str(round(Measures.smape(data, externalforecasts[k][:-1]), 2))+ "		& "
-            ret += str(round(Measures.UStatistic(np.array(data), np.array(forecasts[:-1])), 2))
+            ret += str(round(Measures.UStatistic(data, externalforecasts[k][:-1]), 2))
             ret += "	\\\\ \n"
     return ret
 

tests/pfts.py

@@ -47,9 +47,21 @@ from pyFTS.benchmarks import ProbabilityDistribution as dist
 
 forecasts = pfts1.forecast(taiex_treino)
 
-pmf1 = dist.ProbabilityDistribution("Original",10,[min(taiex_treino),max(taiex_treino)],data=forecasts)
+pmf1 = dist.ProbabilityDistribution("Original",100,[min(taiex_treino),max(taiex_treino)],data=taiex_treino)
 
-print(pmf1)
+#print(pmf1.entropy())
+
+pmf2 = dist.ProbabilityDistribution("Original",100,[min(taiex_treino),max(taiex_treino)],data=forecasts)
+
+#print(pmf2.entropy())
+
+#print(pmf2.kullbackleiblerdivergence(pmf1))
+
+#print(pmf2.crossentropy(pmf1))
+
+print(pmf1.averageloglikelihood(taiex_treino))
+
+print(pmf2.averageloglikelihood(taiex_treino))
 
 #pfts2 = pfts.ProbabilisticFTS("n = 2")
 #pfts2.appendTransformation(diff)