source: branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/IPolicy.cs @ 11793

using System;
using System.Collections.Generic;
using System.Dynamic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using HeuristicLab.Problems.GrammaticalOptimization;

namespace HeuristicLab.Algorithms.Bandits {
  // this interface represents a policy for episodic reinforcement learning (with afterstates)
  // here we assume that a reward is only recieved at the end of the episode and the update is done only after an episode is complete
  // we also assume that the policy can fail to select one of the followStates
  public interface IPolicy<in TState> {
    bool TrySelect(Random random, TState curState, IEnumerable<TState> afterStates, out int selectedStateIdx); // selectedState \in afterStates

    // state-trajectory are the states of the episode, at the end we recieved the reward (only for the terminal state)
    void UpdateReward(IEnumerable<TState> stateTrajectory, double reward);

    void Reset(); // clears all internal state

    // for introspection
    double GetValue(TState state); 
    int GetTries(TState state); 
  }
}