source: branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.Bandits/BanditHelper.cs @ 13402

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace HeuristicLab.Algorithms.Bandits {
  class BanditHelper {
    public static void SampleArms(Random random, IBandit bandit, int nSamples, out double expRewardEst, out int bestArmForExpReward, out int bestArmForMaxReward) {
      bestArmForExpReward = 0;
      bestArmForMaxReward = 0;
      expRewardEst = 0.0;
      var bestSumReward = 0.0;
      var bestUpperPercCount = 0;
      for (int a = 0; a < bandit.NumArms; a++) {
        var sumReward = 0.0;
        var upperPercCount = 0;

        for (int i = 0; i < nSamples; i++) {
          var reward = bandit.Pull(a);
          sumReward += reward;
          if (reward >= 0.95) upperPercCount++;
        }

        if (sumReward > bestSumReward) {
          bestSumReward = sumReward;
          bestArmForExpReward = a;
          expRewardEst = bestSumReward / nSamples;
        }
        if (upperPercCount > bestUpperPercCount) {
          bestUpperPercCount = upperPercCount;
          bestArmForMaxReward = a;
        }
      }
    }
  }
}