Changeset 11745 for branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/BanditPolicies

Timestamp:

01/10/15 14:06:29 (10 years ago)

Author:

gkronber

Message:

#2283: worked on contextual MCTS

File:

• branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/BanditPolicies/UCB1Policy.cs (modified) (1 diff)

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/BanditPolicies/UCB1Policy.cs

@@ -10 +10 @@
   public class UCB1Policy : IBanditPolicy {
     public int SelectAction(Random random, IEnumerable<IBanditPolicyActionInfo> actionInfos) {
-      var myActionInfos = actionInfos.OfType<DefaultPolicyActionInfo>().ToArray(); // TODO: performance
+      var myActionInfos = actionInfos.OfType<DefaultPolicyActionInfo>();
       int bestAction = -1;
       double bestQ = double.NegativeInfinity;
       int totalTries = myActionInfos.Where(a => !a.Disabled).Sum(a => a.Tries);
 
-      for (int a = 0; a < myActionInfos.Length; a++) {
-        if (myActionInfos[a].Disabled) continue;
-        if (myActionInfos[a].Tries == 0) return a;
-        var q = myActionInfos[a].SumReward / myActionInfos[a].Tries + Math.Sqrt((2 * Math.Log(totalTries)) / myActionInfos[a].Tries);
+      int aIdx = -1;
+      foreach (var aInfo in myActionInfos) {
+        aIdx++;
+        if (aInfo.Disabled) continue;
+        if (aInfo.Tries == 0) return aIdx;
+        var q = aInfo.SumReward / aInfo.Tries + Math.Sqrt((2 * Math.Log(totalTries)) / aInfo.Tries);
         if (q > bestQ) {
           bestQ = q;
-          bestAction = a;
+          bestAction = aIdx;
         }
       }