training_config.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_RL_TRAINING_CONFIG_HPP
#define MLPACK_METHODS_RL_TRAINING_CONFIG_HPP

namespace mlpack {
namespace rl {

class TrainingConfig
{
 public:
  TrainingConfig() :
      numWorkers(1),
      updateInterval(1),
      targetNetworkSyncInterval(100),
      stepLimit(0),
      explorationSteps(1),
      stepSize(0.01),
      discount(0.99),
      gradientLimit(40),
      doubleQLearning(false),
      noisyQLearning(false)
  { /* Nothing to do here. */ }

  TrainingConfig(
      size_t numWorkers,
      size_t updateInterval,
      size_t targetNetworkSyncInterval,
      size_t stepLimit,
      size_t explorationSteps,
      double stepSize,
      double discount,
      double gradientLimit,
      bool doubleQLearning,
      bool noisyQLearning) :
      numWorkers(numWorkers),
      updateInterval(updateInterval),
      targetNetworkSyncInterval(targetNetworkSyncInterval),
      stepLimit(stepLimit),
      explorationSteps(explorationSteps),
      stepSize(stepSize),
      discount(discount),
      gradientLimit(gradientLimit),
      doubleQLearning(doubleQLearning),
      noisyQLearning(noisyQLearning)
  { /* Nothing to do here. */ }

  size_t NumWorkers() const { return numWorkers; }
  size_t& NumWorkers() { return numWorkers; }

  size_t UpdateInterval() const { return updateInterval; }
  size_t& UpdateInterval() { return updateInterval; }

  size_t TargetNetworkSyncInterval() const
  { return targetNetworkSyncInterval; }
  size_t& TargetNetworkSyncInterval() { return targetNetworkSyncInterval; }

  size_t StepLimit() const { return stepLimit; }
  size_t& StepLimit() { return stepLimit; }

  size_t ExplorationSteps() const { return explorationSteps; }
  size_t& ExplorationSteps() { return explorationSteps; }

  double StepSize() const { return stepSize; }
  double& StepSize() { return stepSize; }

  double Discount() const { return discount; }
  double& Discount() { return discount; }

  double GradientLimit() const { return gradientLimit; }
  double& GradientLimit() { return gradientLimit; }

  bool DoubleQLearning() const { return doubleQLearning; }
  bool& DoubleQLearning() { return doubleQLearning; }

  bool NoisyQLearning() const { return noisyQLearning; }
  bool& NoisyQLearning() { return noisyQLearning; }

 private:
  size_t numWorkers;

  size_t updateInterval;

  size_t targetNetworkSyncInterval;

  size_t stepLimit;

  size_t explorationSteps;

  double stepSize;

  double discount;

  double gradientLimit;

  bool doubleQLearning;

  bool noisyQLearning;
};

} // namespace rl
} // namespace mlpack

#endif