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 <limits>

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 = std::numeric_limits<size_t>::max());

  template<typename InputType, typename OutputType>
  void Forward(InputType&& input, OutputType&& output);

  template<typename InputType, typename ErrorType, typename GradientType>
  void Backward(const InputType&& input,
                ErrorType&& gy,
                GradientType&& g);

  /*
   * Reset the layer parameter.
   */
  void Reset();

  /*
   * Resets the cell to accept a new input. This breaks the BPTT chain starts a
   * new one.
   *
   * @param size The current maximum number of steps through time.
   */
  void ResetCell(const size_t size);

  /*
   * 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 InputType, typename ErrorType, typename GradientType>
  void Gradient(InputType&& input,
                ErrorType&& error,
                GradientType&& gradient);

  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 grad; }
  OutputDataType& Gradient() { return grad; }

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

 private:
  size_t inSize;

  size_t outSize;

  size_t rho;

  size_t forwardStep;

  size_t backwardStep;

  size_t gradientStep;

  OutputDataType weights;

  OutputDataType prevOutput;

  size_t batchSize;

  size_t batchStep;

  size_t gradientStepIdx;

  OutputDataType cellActivationError;

  OutputDataType delta;

  OutputDataType grad;

  InputDataType inputParameter;

  OutputDataType outputParameter;

  OutputDataType output2GateInputWeight;

  OutputDataType input2GateInputWeight;

  OutputDataType input2GateInputBias;

  OutputDataType cell2GateInputWeight;

  OutputDataType output2GateForgetWeight;

  OutputDataType input2GateForgetWeight;

  OutputDataType input2GateForgetBias;

  OutputDataType cell2GateForgetWeight;

  OutputDataType output2GateOutputWeight;

  OutputDataType input2GateOutputWeight;

  OutputDataType input2GateOutputBias;

  OutputDataType cell2GateOutputWeight;

  OutputDataType inputGate;

  OutputDataType forgetGate;

  OutputDataType hiddenLayer;

  OutputDataType outputGate;

  OutputDataType inputGateActivation;

  OutputDataType forgetGateActivation;

  OutputDataType outputGateActivation;

  OutputDataType hiddenLayerActivation;

  OutputDataType input2HiddenWeight;

  OutputDataType input2HiddenBias;

  OutputDataType output2HiddenWeight;

  OutputDataType cell;

  OutputDataType cellActivation;

  OutputDataType forgetGateError;

  OutputDataType outputGateError;

  OutputDataType prevError;

  OutputDataType outParameter;

  OutputDataType inputCellError;

  OutputDataType inputGateError;

  OutputDataType hiddenError;

  size_t rhoSize;

  size_t bpttSteps;
}; // class LSTM

} // namespace ann
} // namespace mlpack

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

#endif