source: branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/Policies/GenericThompsonSamplingPolicy.cs @ 11730

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using HeuristicLab.Common;

namespace HeuristicLab.Algorithms.Bandits {
  public class GenericThompsonSamplingPolicy : BanditPolicy {
    private readonly Random random;
    private readonly IModel model;

    public GenericThompsonSamplingPolicy(Random random, int numActions, IModel model)
      : base(numActions) {
      this.random = random;
      this.model = model;
    }

    public override int SelectAction() {
      Debug.Assert(Actions.Any());
      var maxR = double.NegativeInfinity;
      int bestAction = -1;
      var expRewards = model.SampleExpectedRewards(random);
      foreach (var a in Actions) {
        var r = expRewards[a];
        if (r > maxR) {
          maxR = r;
          bestAction = a;
        }
      }
      return bestAction;
    }

    public override void UpdateReward(int action, double reward) {
      Debug.Assert(Actions.Contains(action));

      model.Update(action, reward);
    }

    public override void DisableAction(int action) {
      base.DisableAction(action);
      model.Disable(action);
    }

    public override void Reset() {
      base.Reset();
      model.Reset();
    }

    public override void PrintStats() {
      model.PrintStats();
    }

    public override string ToString() {
      return string.Format("GenericThompsonSamplingPolicy({0})", model);
    }
  }
}