lstm.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LAYER_LSTM_HPP
#define MLPACK_METHODS_ANN_LAYER_LSTM_HPP

#include <mlpack/prereqs.hpp>

#include <boost/ptr_container/ptr_vector.hpp>

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

#include "layer_types.hpp"
#include "add_merge.hpp"
#include "sequential.hpp"

namespace mlpack {
namespace ann  {

template <
    typename InputDataType = arma::mat,
    typename OutputDataType = arma::mat
>
class LSTM
{
 public:
  LSTM();

  LSTM(const size_t inSize, const size_t outSize, const size_t rho);

  template<typename eT>
  void Forward(arma::Mat<eT>&& input, arma::Mat<eT>&& output);

  template<typename eT>
  void Backward(const arma::Mat<eT>&& /* input */,
                arma::Mat<eT>&& gy,
                arma::Mat<eT>&& g);

  /*
   * Calculate the gradient using the output delta and the input activation.
   *
   * @param input The input parameter used for calculating the gradient.
   * @param error The calculated error.
   * @param gradient The calculated gradient.
   */
  template<typename eT>
  void Gradient(arma::Mat<eT>&& input,
                arma::Mat<eT>&& /* error */,
                arma::Mat<eT>&& /* gradient */);

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

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

  OutputDataType const& Parameters() const { return weights; }
  OutputDataType& Parameters() { return weights; }

  InputDataType const& InputParameter() const { return inputParameter; }
  InputDataType& InputParameter() { return inputParameter; }

  OutputDataType const& OutputParameter() const { return outputParameter; }
  OutputDataType& OutputParameter() { return outputParameter; }

  OutputDataType const& Delta() const { return delta; }
  OutputDataType& Delta() { return delta; }

  OutputDataType const& Gradient() const { return gradient; }
  OutputDataType& Gradient() { return gradient; }

  std::vector<LayerTypes>& Model() { return network; }

  template<typename Archive>
  void Serialize(Archive& ar, const unsigned int /* version */);

 private:

  size_t inSize;

  size_t outSize;

  size_t rho;

  OutputDataType weights;

  arma::mat prevOutput;

  arma::mat prevCell;

  LayerTypes input2GateModule;

  LayerTypes output2GateModule;

  LayerTypes inputGateModule;

  LayerTypes hiddenStateModule;

  LayerTypes forgetGateModule;

  LayerTypes outputGateModule;

  LayerTypes cellModule;

  LayerTypes cellActivationModule;

  OutputParameterVisitor outputParameterVisitor;

  DeltaVisitor deltaVisitor;

  std::vector<LayerTypes> network;

  size_t forwardStep;

  size_t backwardStep;

  size_t gradientStep;

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

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

  arma::mat prevError;

  arma::mat cellActivationError;

  arma::mat forgetGateError;

  bool deterministic;

  OutputDataType delta;

  OutputDataType gradient;

  InputDataType inputParameter;

  OutputDataType outputParameter;
}; // class LSTM

} // namespace ann
} // namespace mlpack

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

#endif