Changeset 12876 for branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Algorithms.Bandits/Policies/UCB1Policy.cs

Timestamp:

08/17/15 19:13:19 (8 years ago)

Author:

gkronber

Message:

#2283: implemented first crude version of extreme hunter algorithm in branch

File:

• branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Algorithms.Bandits/Policies/UCB1Policy.cs (modified) (2 diffs)

branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Algorithms.Bandits/Policies/UCB1Policy.cs

@@ -10 +10 @@
   // policy for k-armed bandit (see Auer et al. 2002)
   public class UCB1Policy : IBanditPolicy {
+    public double MaxReward { get; private set; }
+    public UCB1Policy(double maxReward = 1.0) {
+      this.MaxReward = maxReward;
+    }
     public int SelectAction(Random random, IEnumerable<IBanditPolicyActionInfo> actionInfos) {
       var myActionInfos = actionInfos.OfType<DefaultPolicyActionInfo>();
@@ -24 +28 @@
         } else {
 
-          q = aInfo.SumReward / aInfo.Tries + 0.5 * Math.Sqrt((2 * Math.Log(totalTries)) / aInfo.Tries);
+          q = aInfo.SumReward / aInfo.Tries + MaxReward * Math.Sqrt((2 * Math.Log(totalTries)) / aInfo.Tries);
         }
         if (q > bestQ) {