From 300678156cb960c5f93bcc618aad273ff3db7790 Mon Sep 17 00:00:00 2001
From: George Lacey <git@george.ooo>
Date: Fri, 13 Oct 2017 16:20:58 +0100
Subject: [PATCH] Implement tournament selection

---
 src/individual.py |  3 +--
 src/lifecycle.py  | 14 +++++++++++---
 src/main.py       | 15 +++++++--------
 src/population.py | 38 +++++++++++++++++++++++++-------------
 4 files changed, 44 insertions(+), 26 deletions(-)

diff --git a/src/individual.py b/src/individual.py
index c1e35b2..9beb145 100644
--- a/src/individual.py
+++ b/src/individual.py
@@ -33,8 +33,7 @@ class Individual(object):
         return one, two
 
     def mutate(self):
-        rand.randint(1, 10)
-        if rand.randint(1, 10) % 2 == 0:
+        if rand.random() < 0.5:
             self.x = self.get_rand_param()
         else:
             self.y = self.get_rand_param()
diff --git a/src/lifecycle.py b/src/lifecycle.py
index 0f71a25..d59805f 100644
--- a/src/lifecycle.py
+++ b/src/lifecycle.py
@@ -9,14 +9,21 @@ class Lifecycle(object):
         self.population = Population(self.params["population_size"])
         self.best_fit = list()
         self.average_fit = list()
+        self.iterations = 0
 
     def start(self):
-        for epoch in range(0, self.params["iter"]):
+        for epoch in range(0, self.params["epochs"]):
+            self.iterations += 1
             elite = self.params["elite"]
             crossover = self.params["crossover"]
-            self.population.advance_generation(elite, crossover)
+            self.population.advance_generation(elite, crossover_rate=crossover, n_arena=4)
             self.best_fit.append(self.population.best_fitness())
             self.average_fit.append(self.population.avg_fitness())
+            if epoch > 50:
+                recent_best = self.best_fit[-50:]
+                if max(recent_best) - min(recent_best) < 0.02:
+                    break
+
 
     def best_fitness(self):
         return self.population.best_fitness()
@@ -33,4 +40,5 @@ class Lifecycle(object):
     def get_csv(self):
         id = {'id': self.id}
         best_fitness = {'best_fit': self.best_fitness()}
-        return {**id, **best_fitness, **self.params}
+        iter = {'iter': self.iterations}
+        return {**id, **best_fitness, **self.params, **iter}
diff --git a/src/main.py b/src/main.py
index d98859d..aadcda6 100644
--- a/src/main.py
+++ b/src/main.py
@@ -8,14 +8,15 @@ import time
 param_example = {'population_size': 10,
                  'elite': round(0.1 * 10),
                  'crossover': round(0.6 * 10),
-                 'iter': 1000}
+                 'epochs': 1000,
+                 'iter': 0}
 
 
-def gen_param(pop, elite, crossover, iter):
+def gen_param(pop, elite, crossover, epochs):
     return {'population_size': pop,
             'elite': round(elite * pop),
             'crossover': round(crossover * pop),
-            'iter': iter}
+            'epochs': epochs}
 
 
 outputs = Queue()
@@ -39,9 +40,9 @@ instances = list()
 
 id = 0
 
-for i in [x * 0.01 for x in range(20, 80, 5)]:
+for i in [x * 1 for x in range(0, 100)]:
     instances.append(Process(name="Thread-%d" % i, target=run_instance,
-                            args=[Lifecycle(id, gen_param(20, 0.1, i, 50)), outputs]))
+                             args=[Lifecycle(id, gen_param(50, 0, 0.5, 500)), outputs]))
     id += 1
 
 start_time = time.time()
@@ -50,7 +51,7 @@ for instance in instances:
     instance.start()
 
 while len([t for t in instances if t.is_alive()]) != 0:
-    print("here")
+    print("Processes left: %d" % len([t for t in instances if t.is_alive()]))
     sleep(1)
 
 end_time = time.time()
@@ -61,5 +62,3 @@ while not outputs.empty():
 output_file.close()
 
 print("Execution took %f seconds" % (end_time - start_time))
-
-#ga.generate_graph(show=True)
diff --git a/src/population.py b/src/population.py
index 1589df7..cf7ad8e 100644
--- a/src/population.py
+++ b/src/population.py
@@ -72,7 +72,7 @@ class Population(object):
 
         return sorted_members
 
-    def advance_generation(self, n_elite, n_crossover):
+    def advance_generation(self, n_elite, crossover_rate=0.5, n_arena=4):
         new_generation = list()
 
         # elitism
@@ -80,25 +80,37 @@ class Population(object):
             new_generation.append(member)
 
         # parent
-        for ind in range(0, n_crossover):
-            parent_one = self.roulette()
-            parent_two = self.roulette()
-            x, y = parent_one.crossover(parent_two)
-            new_generation.append(x)
-            new_generation.append(y)
+        # for ind in range(0, n_crossover):
+            # self.roulette_crossover(new_generation)
 
         while len(new_generation) < len(self.members):
-            new_generation.append(self.roulette())
+            if n_arena > 0:
+                x, y = self.tournament_selection(n_arena, crossover_rate)
+            else:
+                x, y = self.roulette_crossover()
+
+            new_generation.append(x)
+            new_generation.append(y)
 
         self.members = new_generation
 
         self.mutate(10)
 
-        self.check_termination()
-
     def remove_member(self, member):
         self.members.remove(member)
 
-    def check_termination(self):
-        # todo: complete
-        pass
+    def roulette_crossover(self):
+        parent_one = self.roulette()
+        parent_two = self.roulette()
+        return parent_one.crossover(parent_two)
+
+    def tournament_selection(self, arena_size, rate):
+        parents = list()
+        for i in range(arena_size):
+            parents.append(self.members[rand.randint(0, len(self.members) - 1)])
+        parents = sorted(parents, key=lambda x: x.fitness(), reverse=True)
+
+        if rand.random() < rate:
+            return parents[0].crossover(parents[1])
+        else:
+            return parents[0], parents[1]