add epsilon greedy bandit

bin/medley

@@ -52,7 +52,7 @@ def main():
         help="select mechanism for choosing orderings of solvers",
         type=str,
         default="neighbor",
-        choices=["neighbor", "random", "MLP", "thompson", "linear", "preset", "exp3", "knearest", "perfect"]
+        choices=["neighbor", "random", "MLP", "thompson", "linear", "preset", "exp3", "knearest", "perfect", "greedy"]
     )
 
     global_parser.add_argument(
@@ -224,6 +224,8 @@ def main():
         classifier = KNearest(args.k, args.epsilon, args.epsilon_decay, args.time_k)
     elif args.classifier == "perfect":
         classifier = PerfectSelector(args.time_k)
+    elif args.classifier == "greedy":
+        classifier = EpsilonGreedy(args.time_k, args.epsilon)
     else:
         raise RuntimeError("classifier not properly set")
 

medleysolver/classifiers.py

@@ -164,6 +164,48 @@ def update(self, solved_prob, rewards):
         if is_solved(solved_prob.result):
             self.solved.append(solved_prob)
 
+class EpsilonGreedy(ClassifierInterface):
+    def __init__(self, time_k, epsilon):
+        self.epsilon=epsilon
+        self.counts=[0 for _ in SOLVERS]
+        self.values=[0.0 for _ in SOLVERS]
+        self.totals=[0.0 for _ in SOLVERS]
+        self.initialized=False
+        super(EpsilonGreedy, self).__init__(time_k)
+
+    def initialize(self):
+        # nothing to do here
+        return
+
+    def get_ordering(self, point, count, problem):
+        order = list(SOLVERS.keys())
+        explore = np.random.binomial(1, self.epsilon)       
+        if explore==1 and sum(self.values)>0:
+            # return list of solvers sorted by self.values
+            value_order = sorted(list(range(len(SOLVERS))), key=lambda x: self.values[x], reverse=True)  
+            order = [list(SOLVERS.keys())[int(choice)] for choice in value_order]
+            return order
+        else:
+            # shuffle to explore
+            np.random.shuffle(order)
+            return order
+
+    def update(self, solved_prob, rewards):
+        for i,r in enumerate(rewards):
+            if r > 0:
+                self.totals[i] += r
+                self.counts[i] += 1
+                self.values[i] = self.totals[i]/self.counts[i]
+            elif r == 0:
+                self.counts[i] += 1
+                self.values[i] = self.totals[i]/self.counts[i]
+            else:
+                pass          
+        if is_solved(solved_prob.result):
+            self.solved.append(solved_prob)
+
+
+
 class LinearBandit(ClassifierInterface):
     def __init__(self, time_k, alpha=2.358):
         self.initialized = False