Changeset 12876 for branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Algorithms.Bandits/Policies/ActiveLearningPolicy.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/ActiveLearningPolicy.cs (modified) (2 diffs)

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

@@ -9 +9 @@
 namespace HeuristicLab.Algorithms.Bandits.BanditPolicies {
   public class ActiveLearningPolicy : IBanditPolicy {
+    public double MaxReward { get; private set; }
+    public ActiveLearningPolicy(double maxReward = 1.0) {
+      this.MaxReward = maxReward;
+    }
     public int SelectAction(Random random, IEnumerable<IBanditPolicyActionInfo> actionInfos) {
       var myActionInfos = actionInfos.OfType<DefaultPolicyActionInfo>();
@@ -29 +33 @@
           q = aInfo.SumReward / aInfo.Tries;
           var b = Math.Sqrt(Math.Log(2.0 * k * totalTries / delta) / (2.0 * aInfo.Tries));
-          u = q + 0.5 * b;
-          l = q - 0.5 * b;
+          u = q + MaxReward * b;
+          l = q - MaxReward * b;
         }
         bestActions.Add(aIdx);