q_learning.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_RL_Q_LEARNING_HPP
#define MLPACK_METHODS_RL_Q_LEARNING_HPP

#include <mlpack/prereqs.hpp>

#include "replay/random_replay.hpp"
#include "replay/prioritized_replay.hpp"
#include "training_config.hpp"

namespace mlpack {
namespace rl {

template <
  typename EnvironmentType,
  typename NetworkType,
  typename UpdaterType,
  typename PolicyType,
  typename ReplayType = RandomReplay<EnvironmentType>
>
class QLearning
{
 public:
  using StateType = typename EnvironmentType::State;

  using ActionType = typename EnvironmentType::Action;

  QLearning(TrainingConfig config,
            NetworkType network,
            PolicyType policy,
            ReplayType replayMethod,
            UpdaterType updater = UpdaterType(),
            EnvironmentType environment = EnvironmentType());

  ~QLearning();

  double Step();

  double Episode();

  const size_t& TotalSteps() const { return totalSteps; }

  StateType& State() { return state; }
  const StateType& State() const { return state; }

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

  bool& Deterministic() { return deterministic; }
  const bool& Deterministic() const { return deterministic; }

  const NetworkType& Network() const { return learningNetwork; }
  NetworkType& Network() { return learningNetwork; }

 private:
  arma::Col<size_t> BestAction(const arma::mat& actionValues);

  TrainingConfig config;

  NetworkType learningNetwork;

  NetworkType targetNetwork;

  UpdaterType updater;
  #if ENS_VERSION_MAJOR >= 2
  typename UpdaterType::template Policy<arma::mat, arma::mat>* updatePolicy;
  #endif

  PolicyType policy;

  ReplayType replayMethod;

  EnvironmentType environment;

  size_t totalSteps;

  StateType state;

  bool deterministic;
};

} // namespace rl
} // namespace mlpack

// Include implementation
#include "q_learning_impl.hpp"
#endif