Merge pull request #757 from Axelrod-Python/mwe

Meta Winner Ensemble

axelrod/strategies/__init__.py

@@ -10,13 +10,14 @@
     MetaPlayer, MetaMajority, MetaMinority, MetaWinner, MetaHunter,
     MetaMajorityMemoryOne, MetaWinnerMemoryOne, MetaMajorityFiniteMemory,
     MetaWinnerFiniteMemory, MetaMajorityLongMemory, MetaWinnerLongMemory,
-    MetaMixer
+    MetaMixer, MetaWinnerEnsemble
     )
 
 all_strategies.extend([MetaHunter, MetaMajority, MetaMinority, MetaWinner,
                        MetaMajorityMemoryOne, MetaWinnerMemoryOne,
                        MetaMajorityFiniteMemory, MetaWinnerFiniteMemory,
-                       MetaMajorityLongMemory, MetaWinnerLongMemory, MetaMixer])
+                       MetaMajorityLongMemory, MetaWinnerLongMemory, MetaMixer,
+                       MetaWinnerEnsemble])
 
 
 # Distinguished strategy collections in addition to

axelrod/strategies/meta.py

@@ -1,4 +1,4 @@
-from axelrod import Actions, Player, obey_axelrod
+from axelrod import Actions, Player, obey_axelrod, random_choice
 from ._strategies import all_strategies
 from .hunter import (
     DefectorHunter, AlternatorHunter, RandomHunter, MathConstantHunter,
@@ -54,7 +54,8 @@ def strategy(self, opponent):
         # Get the results of all our players.
         results = [player.strategy(opponent) for player in self.team]
 
-        # A subclass should just define a way to choose the result based on team results.
+        # A subclass should just define a way to choose the result based on
+        # team results.
         return self.meta_strategy(results, opponent)
 
     def meta_strategy(self, results, opponent):
@@ -137,7 +138,8 @@ def __init__(self, team=None):
             t.score = 0
 
     def strategy(self, opponent):
-        # Update the running score for each player, before determining the next move.
+        # Update the running score for each player, before determining the
+        # next move.
         if len(self.history):
             for player in self.team:
                 game = self.match_attributes["game"]
@@ -149,7 +151,8 @@ def strategy(self, opponent):
     def meta_strategy(self, results, opponent):
         scores = [pl.score for pl in self.team]
         bestscore = max(scores)
-        beststrategies = [i for i, pl in enumerate(self.team) if pl.score == bestscore]
+        beststrategies = [i for (i, pl) in enumerate(self.team)
+                          if pl.score == bestscore]
         bestproposals = [results[i] for i in beststrategies]
         bestresult = C if C in bestproposals else D
 
@@ -165,6 +168,38 @@ def meta_strategy(self, results, opponent):
 
         return bestresult
 
+class MetaWinnerEnsemble(MetaWinner):
+    """A variant of MetaWinner that chooses one of the top scoring strategies
+    at random against each opponent.
+
+    Names:
+
+    Meta Winner Ensemble: Original name by Marc Harper
+    """
+
+    name = "Meta Winner Ensemble"
+
+    def meta_strategy(self, results, opponent):
+        # Sort by score
+        scores = [(pl.score, i) for (i, pl) in enumerate(self.team)]
+        # Choose one of the best scorers at random
+        scores.sort(reverse=True)
+        prop = max(1, int(len(scores) * 0.08))
+        index = choice([i for (s, i) in scores[:prop]])
+
+        # Update each player's proposed history with his proposed result, but
+        # always after the new result has been settled based on scores
+        # accumulated until now.
+        for r, t in zip(results, self.team):
+            t.proposed_history.append(r)
+
+        if opponent.defections == 0:
+            # Don't poke the bear
+            return C
+
+        # return result
+        return results[index]
+
 
 class MetaHunter(MetaPlayer):
     """A player who uses a selection of hunters."""

axelrod/tests/unit/test_classification.py

@@ -129,6 +129,11 @@ def test_is_basic(self):
             self.assertFalse(axelrod.is_basic(strategy()), msg=strategy)
 
 
+def str_reps(xs):
+    """Maps a collection of player classes to their string representations."""
+    return set(map(str, [x() for x in xs]))
+
+
 class TestStrategies(unittest.TestCase):
 
     def test_strategy_list(self):
@@ -159,39 +164,46 @@ def test_inclusion_of_strategy_lists(self):
                               axelrod.strategies,
                               axelrod.ordinary_strategies,
                               axelrod.cheating_strategies]:
-            self.assertTrue(set(strategy_list).issubset(all_strategies_set))
+            self.assertTrue(str_reps(strategy_list).issubset(
+                str_reps(all_strategies_set)))
 
         strategies_set = set(axelrod.strategies)
         for strategy_list in [axelrod.demo_strategies,
                               axelrod.basic_strategies,
                               axelrod.long_run_time_strategies]:
-            self.assertTrue(set(strategy_list).issubset(strategies_set))
+            self.assertTrue(str_reps(strategy_list).issubset(
+                str_reps(strategies_set)))
 
     def test_long_run_strategies(self):
         long_run_time_strategies = [axelrod.MetaMajority,
                                     axelrod.MetaMinority,
                                     axelrod.MetaWinner,
+                                    axelrod.MetaWinnerEnsemble,
                                     axelrod.MetaMajorityMemoryOne,
                                     axelrod.MetaWinnerMemoryOne,
                                     axelrod.MetaMajorityFiniteMemory,
                                     axelrod.MetaWinnerFiniteMemory,
                                     axelrod.MetaMajorityLongMemory,
                                     axelrod.MetaWinnerLongMemory,
                                     axelrod.MetaMixer]
-        self.assertTrue(long_run_time_strategies,
-                        axelrod.long_run_time_strategies)
+
+        self.assertEqual(str_reps(long_run_time_strategies),
+                         str_reps(axelrod.long_run_time_strategies))
 
     def test_meta_inclusion(self):
-        self.assertTrue(axelrod.MetaMajority in axelrod.strategies)
+        self.assertTrue(str(axelrod.MetaMajority()) in
+                        str_reps(axelrod.strategies))
 
-        self.assertTrue(axelrod.MetaHunter in axelrod.strategies)
-        self.assertFalse(
-            axelrod.MetaHunter in axelrod.long_run_time_strategies)
+        self.assertTrue(str(axelrod.MetaHunter()) in
+                        str_reps(axelrod.strategies))
+        self.assertFalse(str(axelrod.MetaHunter()) in
+            str_reps(axelrod.long_run_time_strategies))
 
     def test_demo_strategies(self):
         demo_strategies = [axelrod.Cooperator,
                            axelrod.Defector,
                            axelrod.TitForTat,
                            axelrod.Grudger,
                            axelrod.Random]
-        self.assertTrue(demo_strategies, axelrod.demo_strategies)
+        self.assertTrue(str_reps(demo_strategies),
+                        str_reps(axelrod.demo_strategies))

axelrod/tests/unit/test_meta.py

@@ -5,7 +5,7 @@
 import axelrod
 import copy
 
-from .test_player import TestPlayer
+from .test_player import TestPlayer, test_responses
 
 C, D = axelrod.Actions.C, axelrod.Actions.D
 
@@ -153,7 +153,8 @@ def test_strategy(self):
         P1 = axelrod.MetaWinner(team=[axelrod.Cooperator, axelrod.Defector])
         P2 = axelrod.Player()
 
-        # This meta player will simply choose the strategy with the highest current score.
+        # This meta player will simply choose the strategy with the highest
+        # current score.
         P1.team[0].score = 0
         P1.team[1].score = 1
         self.assertEqual(P1.strategy(P2), C)
@@ -185,6 +186,35 @@ def test_strategy(self):
         self.assertEqual(player.history[-1], D)
 
 
+class TestMetaWinnerEnsemble(TestMetaPlayer):
+    name = "Meta Winner Ensemble"
+    player = axelrod.MetaWinnerEnsemble
+    expected_classifier = {
+        'memory_depth': float('inf'),  # Long memory
+        'stochastic': True,
+        'makes_use_of': set(['game']),
+        'long_run_time': True,
+        'inspects_source': False,
+        'manipulates_source': False,
+        'manipulates_state': False
+    }
+
+    expected_class_classifier = copy.copy(expected_classifier)
+    expected_class_classifier['stochastic'] = False
+    expected_class_classifier['makes_use_of'] = set([])
+
+    def test_strategy(self):
+        self.first_play_test(C)
+
+        P1 = axelrod.MetaWinner(team=[axelrod.Cooperator, axelrod.Defector])
+        P2 = axelrod.Cooperator()
+        test_responses(self, P1, P2, [C] * 4, [C] * 4, [C] * 4)
+
+        P1 = axelrod.MetaWinner(team=[axelrod.Cooperator, axelrod.Defector])
+        P2 = axelrod.Defector()
+        test_responses(self, P1, P2, [C] * 4, [D] * 4, [D] * 4)
+
+
 class TestMetaHunter(TestMetaPlayer):
 
     name = "Meta Hunter"
@@ -201,7 +231,8 @@ class TestMetaHunter(TestMetaPlayer):
     def test_strategy(self):
         self.first_play_test(C)
 
-        # We are not using the Cooperator Hunter here, so this should lead to cooperation.
+        # We are not using the Cooperator Hunter here, so this should lead to
+        #  cooperation.
         self.responses_test([C, C, C, C], [C, C, C, C], [C])
 
         # After long histories tit-for-tat should come into play.

docs/tutorials/advanced/classification_of_strategies.rst

@@ -101,7 +101,7 @@ Some strategies have been classified as having a particularly long run time::
     ... }
     >>> strategies = axl.filtered_strategies(filterset)
     >>> len(strategies)
-    10
+    11
 
 Strategies that :code:`manipulate_source`, :code:`manipulate_state`
 and/or :code:`inspect_source` return :code:`False` for the :code:`obey_axelrod`