rnn.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_RNN_HPP
#define MLPACK_METHODS_ANN_RNN_HPP

#include <mlpack/prereqs.hpp>

#include "visitor/delete_visitor.hpp"
#include "visitor/delta_visitor.hpp"
#include "visitor/output_parameter_visitor.hpp"
#include "visitor/reset_visitor.hpp"

#include "init_rules/network_init.hpp"

#include <mlpack/methods/ann/layer/layer_types.hpp>
#include <mlpack/methods/ann/layer/layer.hpp>
#include <mlpack/methods/ann/layer/layer_traits.hpp>
#include <mlpack/methods/ann/init_rules/random_init.hpp>

#include <ensmallen.hpp>

namespace mlpack {
namespace ann  {

template<
  typename OutputLayerType = NegativeLogLikelihood<>,
  typename InitializationRuleType = RandomInitialization,
  typename... CustomLayers
>
class RNN
{
 public:
  using NetworkType = RNN<OutputLayerType,
                          InitializationRuleType,
                          CustomLayers...>;

  RNN(const size_t rho,
      const bool single = false,
      OutputLayerType outputLayer = OutputLayerType(),
      InitializationRuleType initializeRule = InitializationRuleType());

  RNN(const RNN&);

  RNN(RNN&&);

  RNN& operator=(const RNN&);

  RNN& operator=(RNN&&);

  ~RNN();

  template<typename OptimizerType>
  typename std::enable_if<
      HasMaxIterations<OptimizerType, size_t&(OptimizerType::*)()>
      ::value, void>::type
  WarnMessageMaxIterations(OptimizerType& optimizer, size_t samples) const;

  template<typename OptimizerType>
  typename std::enable_if<
      !HasMaxIterations<OptimizerType, size_t&(OptimizerType::*)()>
      ::value, void>::type
  WarnMessageMaxIterations(OptimizerType& optimizer, size_t samples) const;

  template<typename OptimizerType, typename... CallbackTypes>
  double Train(arma::cube predictors,
               arma::cube responses,
               OptimizerType& optimizer,
               CallbackTypes&&... callbacks);

  template<typename OptimizerType = ens::StandardSGD, typename... CallbackTypes>
  double Train(arma::cube predictors,
               arma::cube responses,
               CallbackTypes&&... callbacks);

  void Predict(arma::cube predictors,
               arma::cube& results,
               const size_t batchSize = 256);

  double Evaluate(const arma::mat& parameters,
                  const size_t begin,
                  const size_t batchSize,
                  const bool deterministic);

  double Evaluate(const arma::mat& parameters,
                  const size_t begin,
                  const size_t batchSize);

  template<typename GradType>
  double EvaluateWithGradient(const arma::mat& parameters,
                              const size_t begin,
                              GradType& gradient,
                              const size_t batchSize);

  void Gradient(const arma::mat& parameters,
                const size_t begin,
                arma::mat& gradient,
                const size_t batchSize);

  void Shuffle();

  /*
   * Add a new module to the model.
   *
   * @param args The layer parameter.
   */
  template <class LayerType, class... Args>
  void Add(Args... args) { network.push_back(new LayerType(args...)); }

  /*
   * Add a new module to the model.
   *
   * @param layer The Layer to be added to the model.
   */
  void Add(LayerTypes<CustomLayers...> layer) { network.push_back(layer); }

  size_t NumFunctions() const { return numFunctions; }

  const arma::mat& Parameters() const { return parameter; }
  arma::mat& Parameters() { return parameter; }

  const size_t& Rho() const { return rho; }
  size_t& Rho() { return rho; }

  const arma::cube& Responses() const { return responses; }
  arma::cube& Responses() { return responses; }

  const arma::cube& Predictors() const { return predictors; }
  arma::cube& Predictors() { return predictors; }

  void Reset();

  void ResetParameters();

  template<typename Archive>
  void serialize(Archive& ar, const uint32_t /* version */);

 private:
  // Helper functions.
  template<typename InputType>
  void Forward(const InputType& input);

  void ResetCells();

  void Backward();

  template<typename InputType>
  void Gradient(const InputType& input);

  void ResetDeterministic();

  void ResetGradients(arma::mat& gradient);

  size_t rho;

  OutputLayerType outputLayer;

  InitializationRuleType initializeRule;

  size_t inputSize;

  size_t outputSize;

  size_t targetSize;

  bool reset;

  bool single;

  std::vector<LayerTypes<CustomLayers...> > network;

  arma::cube predictors;

  arma::cube responses;

  arma::mat parameter;

  size_t numFunctions;

  arma::mat error;

  DeltaVisitor deltaVisitor;

  OutputParameterVisitor outputParameterVisitor;

  std::vector<arma::mat> moduleOutputParameter;

  WeightSizeVisitor weightSizeVisitor;

  CopyVisitor<CustomLayers...> copyVisitor;

  ResetVisitor resetVisitor;

  DeleteVisitor deleteVisitor;

  bool deterministic;

  arma::mat currentGradient;

  // The BRN class should have access to internal members.
  template<
    typename OutputLayerType1,
    typename MergeLayerType1,
    typename MergeOutputType1,
    typename InitializationRuleType1,
    typename... CustomLayers1
  >
  friend class BRNN;
}; // class RNN

} // namespace ann
} // namespace mlpack

// Include implementation.
#include "rnn_impl.hpp"

#endif