source: branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Algorithms.Bandits/ActionInfos/ModelPolicyActionInfo.cs @ 13777

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

namespace HeuristicLab.Algorithms.Bandits.BanditPolicies {
  // uses a statistical model to sample and update posterior distribution p(Reward | Data)
  public class ModelPolicyActionInfo : IBanditPolicyActionInfo {
    private readonly IModel model;
    public double MaxReward { get; private set; }
    public double Value {
      get {
        return model.Sample(new Random());
      }
    }

    public int Tries { get; private set; }
    public ModelPolicyActionInfo(IModel model) {
      this.model = model;
    }

    public void UpdateReward(double reward) {
      Tries++;
      MaxReward = Math.Max(MaxReward, reward);
      model.Update(reward);
    }

    public double SampleExpectedReward(Random random) {
      return model.Sample(random);
    }

    public void Reset() {
      Tries = 0;
      model.Reset();
    }

    public override string ToString() {
      return string.Format("model {1}", model);
    }
  }
}