Added unique best genomes method to StatisticsReporter.

Added refactoring notes.
Randomly inherit ConnectionGene traits from parents instead of copying one randomly selected parent's gene.

neat/genes.py

@@ -112,9 +112,13 @@ def split(self, innovation_indexer, node_id):
     def copy(self):
         return ConnectionGene(self.innovation_id, self.in_node_id, self.out_node_id, self.weight, self.enabled)
 
-    def is_same_innov(self, cg):
-        return self.innovation_id == cg.innovation_id
+    def is_same_innov(self, other):
+        return self.innovation_id == other.innovation_id
 
-    def get_child(self, cg):
-        # TODO: average both weights (Stanley, p. 38)
-        return random.choice((self, cg)).copy()
+    def get_child(self, other):
+        """ Creates a new ConnectionGene randomly inheriting attributes from its parents."""
+        assert self.innovation_id == other.innovation_id
+        cg = ConnectionGene(self.innovation_id, self.in_node_id, self.out_node_id,
+                      random.choice((self.weight, other.weight)),
+                      random.choice((self.enabled, other.enabled)))
+        return cg

neat/indexer.py

@@ -16,6 +16,10 @@ def clear(self):
         self.next_id = 1
 
 
+# TODO: Why not just use (in, out) as a unique innovation index?  That's what
+# the current scheme ends up doing anyway.  If the scheme needs to be configurable,
+# then the key member of connection genes should be used everywhere instead of
+# having hard-coded (input, output) tuple keys sprinkled all over the code.
 class InnovationIndexer(object):
     def __init__(self, first):
         self.indexer = Indexer(first)

neat/reporting.py

@@ -155,6 +155,18 @@ def get_average_fitness(self):
 
         return avg_fitness
 
+    def best_unique_genomes(self, n):
+        """Returns the most n fit genomes, with no duplication."""
+        best_unique = {}
+        for g in self.most_fit_genomes:
+            best_unique[g.ID] = g
+        best_unique = list(best_unique.values())
+
+        def key(g):
+            return g.fitness
+
+        return sorted(best_unique, key=key, reverse=True)[:n]
+
     def best_genomes(self, n):
         """Returns the n most fit genomes ever seen."""
         def key(g):