implement stagnation

setup.py

@@ -5,7 +5,7 @@
 
 setup(
     name='sneat',
-    version='0.0.6',
+    version='0.0.7',
     packages=find_packages(),
     package_data={'sneat': ['default_config.ini']},
     install_requires=[

sneat/default_config.ini

@@ -23,4 +23,6 @@ remove_node=0.08
 
 [Evolution]
 max_generations = 100
-max_fitness = 4
+max_fitness = 4
+max_stagnation = 15
+min_species = 3

sneat/evolve.py

@@ -38,11 +38,11 @@ def load_checkpoint():
 
 def print_stats(pop):
     print(f'\n\n[i] Gen. {pop.generation}:')
-    headers = ['Species', 'Members', 'Best Fitness', 'Average Fitness']
+    headers = ['Species', 'Members', 'Best Fitness', 'Average Fitness', 'Stagnation', 'Best Complexity']
     for s in pop.species:
         s.members = sorted(s.members, key=lambda x: x.fitness, reverse=True)
     species = sorted(pop.species, key=lambda x: x.members[0].fitness, reverse=True)
-    data = [[s.id, len(s.members), round(max(g.fitness for g in s.members), 2), round(np.mean([g.fitness for g in s.members]), 2)] for s in species]
+    data = [[s.id, len(s.members), round(max(g.fitness for g in s.members), 2), round(np.mean([g.fitness for g in s.members]), 2), s.stagnation, f'{len(s.members[0].network.nodes)}n + {len(s.members[0].network.connections)}'] for s in species]
     print(tb(data, headers=headers))
     print('-' * 55)
 

sneat/population.py

@@ -42,13 +42,30 @@ def reproduce(self):
         for g in self.genomes:
             g.normalized_fitness = (g.fitness - min_fitness) / (max_fitness - min_fitness)
 
-        # assign adjusted fitness scores
         for s in self.species:
-            s_size = len(s.members)
 
+            # bump stagnation
+            if s.members[0].fitness > s.best_fitness:
+                s.best_fitness = s.members[0].fitness
+                s.stagnation = 0
+            else:
+                s.stagnation += 1
+
+            # assign adjusted fitness scores
+            s_size = len(s.members)
             for g in s.members:
                 g.adjusted_fitness = max(g.normalized_fitness / s_size, 0.0001) # avoid division by zero
 
+        # remove stagnant species
+        while len(self.species) >= self.config.getint('Evolution', 'min_species'):
+            stagnant_species = [s for s in self.species if s.stagnation >= self.config.getint('Evolution', 'max_stagnation')]
+            stagnant_species = sorted(stagnant_species, key=lambda x: x.best_fitness, reverse=True)
+            if not stagnant_species:
+                break
+            extinct = stagnant_species.pop()
+            self.species.remove(extinct)
+            print(f'[i] Species {extinct.id} went extinct')
+
         # perform reproduction inside of each species
         offspring = []
         for s in self.species:

sneat/species.py

@@ -4,6 +4,7 @@ def __init__(self, representative, callbacks):
         self.members = [representative]
         self.id = callbacks['get_next_species_id']()
         self.stagnation = 0
+        self.best_fitness = float('-inf')
 
     def add_member(self, genome):
         self.members.append(genome)