Changeset 11732 for branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/Models/BernoulliModel.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/Models/BernoulliModel.cs (modified) (2 diffs)

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/Models/BernoulliModel.cs

@@ -9 +9 @@
 namespace HeuristicLab.Algorithms.Bandits.Models {
   public class BernoulliModel : IModel {
-    private readonly int numActions;
-    private readonly int[] success;
-    private readonly int[] failure;
+    private int success;
+    private int failure;
 
     // parameters of beta prior distribution
@@ -17 +16 @@
     private readonly double beta;
 
-    public BernoulliModel(int numActions, double alpha = 1.0, double beta = 1.0) {
-      this.numActions = numActions;
-      this.success = new int[numActions];
-      this.failure = new int[numActions];
+    public BernoulliModel(double alpha = 1.0, double beta = 1.0) {
       this.alpha = alpha;
       this.beta = beta;
     }
 
-
-    public double[] SampleExpectedRewards(Random random) {
+    public double SampleExpectedReward(Random random) {
       // sample bernoulli mean from beta prior
-      var theta = new double[numActions];
-      for (int a = 0; a < numActions; a++) {
-        if (success[a] == -1)
-          theta[a] = 0.0;
-        else {
-          theta[a] = Rand.BetaRand(random, success[a] + alpha, failure[a] + beta);
-        }
-      }
-
-      // no need to sample we know the exact expected value
-      // the expected value of a bernoulli variable is just theta
-      return theta.Select(t => t).ToArray();
+      return Rand.BetaRand(random, success + alpha, failure + beta);
     }
 
-    public void Update(int action, double reward) {
-      const double EPSILON = 1E-6;
-      Debug.Assert(Math.Abs(reward - 0.0) < EPSILON || Math.Abs(reward - 1.0) < EPSILON);
-      if (Math.Abs(reward - 1.0) < EPSILON) {
-        success[action]++;
+    public void Update(double reward) {
+      Debug.Assert(reward.IsAlmost(1.0) || reward.IsAlmost(0.0));
+      if (reward.IsAlmost(1.0)) {
+        success++;
       } else {
-        failure[action]++;
+        failure++;
       }
     }
 
-    public void Disable(int action) {
-      success[action] = -1;
-    }
-
     public void Reset() {
-      Array.Clear(success, 0, numActions);
-      Array.Clear(failure, 0, numActions);
+      success = 0;
+      failure = 0;
     }
 
     public void PrintStats() {
-      for (int i = 0; i < numActions; i++) {
-        Console.Write("{0:F2} ", success[i] / (double)failure[i]);
-      }
+      Console.Write("{0:F2} ", success / (double)failure);
+    }
+
+    public object Clone() {
+      return new BernoulliModel() { failure = this.failure, success = this.success };
     }
   }