reward_clipping.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_RL_ENVIRONMENT_REWARD_CLIPPING_HPP
#define MLPACK_METHODS_RL_ENVIRONMENT_REWARD_CLIPPING_HPP

#include <mlpack/prereqs.hpp>
#include <mlpack/core/math/clamp.hpp>

namespace mlpack {
namespace rl {

template <typename EnvironmentType>
class RewardClipping
{
 public:
  using State = typename EnvironmentType::State;

  using Action = typename EnvironmentType::Action;

  RewardClipping(EnvironmentType& environment,
                 const double minReward = -1.0,
                 const double maxReward = 1.0) :
    environment(environment),
    minReward(minReward),
    maxReward(maxReward)
  {
    // Nothing to do here
  }

  State InitialSample()
  {
    return environment.InitialSample();
  }

  bool IsTerminal(const State& state) const
  {
    return environment.IsTerminal(state);
  }

  double Sample(const State& state,
                const Action& action,
                State& nextState)
  {
    // Get original unclipped reward from base environment.
    double unclippedReward =  environment.Sample(state, action, nextState);
    // Clip rewards according to the min and max limit and return.
    return math::ClampRange(unclippedReward, minReward, maxReward);
  }

  double Sample(const State& state, const Action& action)
  {
    State nextState;
    return Sample(state, action, nextState);
  }

  EnvironmentType& Environment() const { return environment; }
  EnvironmentType& Environment() { return environment; }

  double MinReward() const { return minReward; }
  double& MinReward() { return minReward; }

  double MaxReward() const { return maxReward; }
  double& MaxReward() { return maxReward; }

 private:
  EnvironmentType environment;

  double minReward;

  double maxReward;
};

} // namespace rl
} // namespace mlpack

#endif