Changeset 11732 for branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/IPolicy.cs

Timestamp:

01/07/15 09:21:46 (9 years ago)

Author:

gkronber

Message:

#2283: refactoring and bug fixes

File:

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

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/IPolicy.cs

@@ -8 +8 @@
   // this interface represents a policy for reinforcement learning
   public interface IPolicy {
-    IEnumerable<int> Actions { get; }
-    int SelectAction(); // action selection ...
-    void UpdateReward(int action, double reward); // ... and reward update are defined as usual
-
-    // policies must also support disabling of potential actions
-    // for instance if we know that an action in a state has a deterministic 
-    // reward we need to sample it only once
-    // it is necessary to sample an action only once
-    void DisableAction(int action);
-
-    // reset causes the policy to be reinitialized to it's initial state (as after constructor-call)
-    void Reset();
-
-    void PrintStats();
+    int SelectAction(Random random, IEnumerable<IPolicyActionInfo> actionInfos);
+    IPolicyActionInfo CreateActionInfo();
   }
 }