Add state to action analysis.

axelrod/interaction_utils.py

@@ -2,7 +2,7 @@
 Functions to calculate results from interactions. Interactions are lists of the
 form:
 
-    [('C', 'D'), ('D', 'C'),...]
+    [(C, D), (D, C),...]
 
 This is used by both the Match class and the ResultSet class which analyse
 interactions.
@@ -135,9 +135,96 @@ def compute_normalised_state_distribution(interactions):
     return normalized_count
 
 
+def compute_state_to_action_distribution(interactions):
+    """
+    Returns a list (for each player) of counts of each state to action pair
+    for a set of interactions. A state to action pair is of the form:
+
+    ((C, D), C)
+
+    Implying that from a state of (C, D) (the first player having played C and
+    the second playing D) the player in question then played C.
+
+    The following counter object implies that the player in question was in
+    state (C, D) for a total of 12 times, subsequently cooperating 4 times and
+    defecting 8 times.
+
+    Counter({((C, D), C): 4, ((C, D), D): 8})
+
+    Parameters
+    ----------
+    interactions : list of tuples
+        A list containing the interactions of the match as shown at the top of
+        this file.
+
+    Returns
+    ----------
+    state_to_C_distributions : List of Counter Object
+        List of Counter objects where the keys are the states and actions and
+        the values the counts. The
+        first/second Counter corresponds to the first/second player.
+    """
+    if not interactions:
+        return None
+
+    distributions = [Counter([(state, outcome[j])
+                     for state, outcome in zip(interactions, interactions[1:])])
+                     for j in range(2)]
+    return distributions
+
+
+def compute_normalised_state_to_action_distribution(interactions):
+    """
+    Returns a list (for each player) of normalised counts of each state to action
+    pair for a set of interactions. A state to action pair is of the form:
+
+    ((C, D), C)
+
+    implying that from a state of (C, D) (the first player having played C and
+    the second playing D) the player in question then played C.
+
+    The following counter object, implies that the player in question was only
+    ever in state (C, D), subsequently cooperating 1/3 of the time and defecting
+    2/3 times.
+
+    Counter({((C, D), C): 0.333333, ((C, D), D): 0.66666667})
+
+    Parameters
+    ----------
+    interactions : list of tuples
+        A list containing the interactions of the match as shown at the top of
+        this file.
+
+    Returns
+    -------
+    normalised_state_to_C_distributions : List of Counter Object
+        List of Counter objects where the keys are the states and actions and
+        the values the normalized counts. The first/second Counter corresponds
+        to the first/second player.
+    """
+    if not interactions:
+        return None
+
+    distribution = compute_state_to_action_distribution(interactions)
+    normalized_distribution = []
+    for player in range(2):
+        counter = {}
+        for state in [(C, C), (C, D), (D, C), (D, D)]:
+            C_count = distribution[player].get((state, C), 0)
+            D_count = distribution[player].get((state, D), 0)
+            total = C_count + D_count
+            if total > 0:
+                if C_count > 0:
+                    counter[(state, C)] = C_count / (C_count + D_count)
+                if D_count > 0:
+                    counter[(state, D)] = D_count / (C_count + D_count)
+        normalized_distribution.append(Counter(counter))
+    return normalized_distribution
+
+
 def sparkline(actions, c_symbol='█', d_symbol=' '):
     return ''.join([
-        c_symbol if play == 'C' else d_symbol for play in actions])
+        c_symbol if play == C else d_symbol for play in actions])
 
 
 def compute_sparklines(interactions, c_symbol='█', d_symbol=' '):

axelrod/result_set.py

@@ -340,8 +340,9 @@ def _build_ranking(self):
     @update_progress_bar
     def _build_normalised_state_distribution(self):
         """
-        Returns
-        ----------
+        Returns:
+        --------
+            norm : list
 
             Normalised state distribution. A list of lists of counter objects:
 
@@ -358,6 +359,34 @@ def _build_normalised_state_distribution(self):
             norm.append(counters)
         return norm
 
+    @update_progress_bar
+    def _build_normalised_state_to_action_distribution(self):
+        """
+        Returns:
+        --------
+            norm : list
+
+            A list of lists of counter objects.
+
+            Dictionary where the keys are the states and the values are a
+            normalized counts of the number of times that state goes to a given
+            action.
+        """
+        norm = []
+        for player in self.state_to_action_distribution:
+            counters = []
+            for counter in player:
+                norm_counter = Counter()
+                for state in [(C, C), (C, D), (D, C), (D, D)]:
+                    total = counter[(state, C)] + counter[(state, D)]
+                    if total > 0:
+                        for action in [C, D]:
+                            if counter[(state, action)] > 0:
+                                norm_counter[(state, action)] = counter[(state, action)] / total
+                counters.append(norm_counter)
+            norm.append(counters)
+        return norm
+
     def _build_empty_metrics(self, keep_interactions=False):
         """
         Creates the various empty metrics ready to be updated as the data is
@@ -385,6 +414,8 @@ def _build_empty_metrics(self, keep_interactions=False):
         self.initial_cooperation_count = [0 for player in plist]
         self.state_distribution = [[Counter() for opponent in plist]
                                    for player in plist]
+        self.state_to_action_distribution = [[Counter() for opponent in plist]
+                                             for player in plist]
         self.good_partner_matrix = [[0 for opponent in plist]
                                     for player in plist]
 
@@ -567,6 +598,26 @@ def _update_state_distribution(self, p1, p2, counter):
         counter[(C, D)], counter[(D, C)] = counter[(D, C)], counter[(C, D)]
         self.state_distribution[p2][p1] += counter
 
+    def _update_state_to_action_distribution(self, p1, p2, counter_list):
+        """
+        During a read of the data, update the state_distribution attribute
+
+        Parameters
+        ----------
+
+            p1, p2 : int
+                The indices of the first and second player
+            counter_list : list of collections.Counter
+                A list of counter objects for the states to action of a match
+        """
+        counter = counter_list[0]
+        self.state_to_action_distribution[p1][p2] += counter
+
+        counter = counter_list[1]
+        for act in [C, D]:
+            counter[((C, D), act)], counter[((D, C), act)] = counter[((D, C), act)], counter[((C, D), act)]
+        self.state_to_action_distribution[p2][p1] += counter
+
     def _update_good_partner_matrix(self, p1, p2, cooperations):
         """
         During a read of the data, update the good partner matrix attribute
@@ -671,6 +722,7 @@ def _build_score_related_metrics(self, progress_bar=False,
                 self._update_normalised_cooperation(p1, p2, interaction)
 
                 if p1 != p2:  # Anything that ignores self interactions
+                    state_to_actions = iu.compute_state_to_action_distribution(interaction)
 
                     for player in [p1, p2]:
                         self.total_interactions[player] += 1
@@ -685,16 +737,19 @@ def _build_score_related_metrics(self, progress_bar=False,
                     self._update_initial_cooperation_count(p1, p2,
                                                            initial_coops)
                     self._update_state_distribution(p1, p2, state_counter)
+                    self._update_state_to_action_distribution(p1, p2,
+                                                              state_to_actions)
                     self._update_good_partner_matrix(p1, p2, cooperations)
 
         if progress_bar:
-            self.progress_bar = tqdm.tqdm(total=12 + 2 * self.nplayers,
+            self.progress_bar = tqdm.tqdm(total=13 + 2 * self.nplayers,
                                           desc="Finishing")
         self._summarise_normalised_scores()
         self._summarise_normalised_cooperation()
 
         self.ranking = self._build_ranking()
         self.normalised_state_distribution = self._build_normalised_state_distribution()
+        self.normalised_state_to_action_distribution = self._build_normalised_state_to_action_distribution()
         self.ranked_names = self._build_ranked_names()
         self.payoff_matrix = self._build_payoff_matrix()
         self.payoff_stddevs = self._build_payoff_stddevs()
@@ -772,7 +827,9 @@ def summarise(self):
         self.player = namedtuple("Player", ["Rank", "Name", "Median_score",
                                             "Cooperation_rating", "Wins",
                                             "Initial_C_rate", "CC_rate",
-                                            "CD_rate", "DC_rate", "DD_rate"])
+                                            "CD_rate", "DC_rate", "DD_rate",
+                                            "CC_to_C_rate", "CD_to_C_rate",
+                                            "DC_to_C_rate", "DD_to_C_rate"])
 
         states = [(C, C), (C, D), (D, C), (D, D)]
         state_prob = []
@@ -787,13 +844,28 @@ def summarise(self):
                 counts = [0 for c in counts]
             state_prob.append(counts)
 
+        state_to_C_prob = []
+        for player in self.normalised_state_to_action_distribution:
+            rates = []
+            for state in states:
+                counts = [counter[(state, 'C')] for counter in player
+                          if counter[(state, 'C')] > 0]
+
+                if len(counts) > 0:
+                    rate = mean(counts)
+                else:
+                    rate = 0
+
+                rates.append(rate)
+            state_to_C_prob.append(rates)
+
         summary_measures = list(zip(self.players, median_scores,
                                     self.cooperating_rating, median_wins,
                                     self.initial_cooperation_rate))
 
         summary_data = []
         for rank, i in enumerate(self.ranking):
-            data = list(summary_measures[i]) + state_prob[i]
+            data = list(summary_measures[i]) + state_prob[i] + state_to_C_prob[i]
             summary_data.append(self.player(rank, *data))
 
         return summary_data
@@ -802,7 +874,8 @@ def write_summary(self, filename):
         """
         Write a csv file containing summary data of the results of the form:
 
-            "Rank", "Name", "Median-score-per-turn", "Cooperation-rating", "Initial_C_Rate", "Wins", "CC-Rate", "CD-Rate", "DC-Rate", "DD-rate"
+            "Rank", "Name", "Median-score-per-turn", "Cooperation-rating", "Initial_C_Rate", "Wins", "CC-Rate", "CD-Rate", "DC-Rate", "DD-rate","CC-to-C-Rate", "CD-to-C-Rate", "DC-to-C-Rate", "DD-to-C-rate"
+
 
         Parameters
         ----------

axelrod/tests/unit/test_interaction_utils.py

@@ -22,11 +22,27 @@ class TestMatch(unittest.TestCase):
                           Counter({('D', 'C'): 2}),
                           Counter({('C', 'C'): 1, ('C', 'D'): 1}),
                           None]
+    state_to_action_distribution = [[Counter({(('C', 'D'), 'D'): 1}),
+                                     Counter({(('C', 'D'), 'C'): 1})],
+                                    [Counter({(('D', 'C'), 'D'): 1}),
+                                     Counter({(('D', 'C'), 'C'): 1})],
+                                    [Counter({(('C', 'C'), 'C'): 1}),
+                                     Counter({(('C', 'C'), 'D'): 1})],
+                                    None]
+
     normalised_state_distribution = [
         Counter({('C', 'D'): 0.5, ('D', 'C'): 0.5}),
         Counter({('D', 'C'): 1.0}),
         Counter({('C', 'C'): 0.5, ('C', 'D'): 0.5}),
         None]
+    normalised_state_to_action_distribution = [[Counter({(('C', 'D'), 'D'): 1}),
+                                                Counter({(('C', 'D'), 'C'): 1})],
+                                               [Counter({(('D', 'C'), 'D'): 1}),
+                                                Counter({(('D', 'C'), 'C'): 1})],
+                                               [Counter({(('C', 'C'), 'C'): 1}),
+                                                Counter({(('C', 'C'), 'D'): 1})],
+                                               None]
+
     sparklines = [ '█ \n █', '  \n██', '██\n█ ', None ]
 
     def test_compute_scores(self):
@@ -63,6 +79,37 @@ def test_compute_normalised_state_distribution(self):
         for inter, dist in zip(self.interactions, self.normalised_state_distribution):
             self.assertEqual(dist, iu.compute_normalised_state_distribution(inter))
 
+    def test_compute_state_to_action_distribution(self):
+        for inter, dist in zip(self.interactions,
+                               self.state_to_action_distribution):
+            self.assertEqual(dist,
+                             iu.compute_state_to_action_distribution(inter))
+        inter = [(C, D), (D, C), (C, D), (D, C), (D, D), (C, C), (C, D)]
+        expected_dist =[Counter({(('C', 'C'), 'C'): 1, (('D', 'C'), 'C'): 1,
+                                 (('C', 'D'), 'D'): 2, (('D', 'C'), 'D'): 1,
+                                 (('D', 'D'), 'C'): 1}),
+                        Counter({(('C', 'C'), 'D'): 1,
+                                 (('C', 'D'), 'C'): 2, (('D', 'C'), 'D'): 2,
+                                 (('D', 'D'), 'C'): 1})]
+
+        self.assertEqual(expected_dist,
+                         iu.compute_state_to_action_distribution(inter))
+
+    def test_compute_normalised_state_to_action_distribution(self):
+        for inter, dist in zip(self.interactions,
+                               self.normalised_state_to_action_distribution):
+            self.assertEqual(dist,
+                             iu.compute_normalised_state_to_action_distribution(inter))
+        inter = [(C, D), (D, C), (C, D), (D, C), (D, D), (C, C), (C, D)]
+        expected_dist =[Counter({(('C', 'C'), 'C'): 1, (('D', 'C'), 'C'): 1 / 2,
+                                 (('C', 'D'), 'D'): 1, (('D', 'C'), 'D'): 1 / 2,
+                                 (('D', 'D'), 'C'): 1}),
+                        Counter({(('C', 'C'), 'D'): 1,
+                                 (('C', 'D'), 'C'): 1, (('D', 'C'), 'D'): 1,
+                                 (('D', 'D'), 'C'): 1})]
+        self.assertEqual(expected_dist,
+                         iu.compute_normalised_state_to_action_distribution(inter))
+
     def test_compute_sparklines(self):
         for inter, spark in zip(self.interactions, self.sparklines):
             self.assertEqual(spark, iu.compute_sparklines(inter))