fast_lstm.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LAYER_FAST_LSTM_HPP
#define MLPACK_METHODS_ANN_LAYER_FAST_LSTM_HPP

#include <mlpack/prereqs.hpp>
#include <limits>

namespace mlpack {
namespace ann  {

template <
    typename InputDataType = arma::mat,
    typename OutputDataType = arma::mat
>
class FastLSTM
{
 public:
  // Convenience typedefs.
  typedef typename InputDataType::elem_type InputElemType;
  typedef typename OutputDataType::elem_type ElemType;

  FastLSTM();

  FastLSTM(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; }

  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:
  template<typename InputType, typename OutputType>
  void FastSigmoid(InputType&& input, OutputType&& sigmoids)
  {
    for (size_t i = 0; i < input.n_elem; ++i)
      sigmoids(i) = FastSigmoid(input(i));
  }

  ElemType FastSigmoid(const InputElemType data)
  {
    ElemType x = 0.5 * data;
    ElemType z;
    if (x >= 0)
    {
      if (x < 1.7)
        z = (1.5 * x / (1 + x));
      else if (x < 3)
        z = (0.935409070603099 + 0.0458812946797165 * (x - 1.7));
      else
        z = 0.99505475368673;
    }
    else
    {
      ElemType xx = -x;
      if (xx < 1.7)
        z = -(1.5 * xx / (1 + xx));
      else if (xx < 3)
        z = -(0.935409070603099 + 0.0458812946797165 * (xx - 1.7));
      else
        z = -0.99505475368673;
    }

    return 0.5 * (z + 1.0);
  }

  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;

  OutputDataType outputParameter;

  OutputDataType output2GateWeight;

  OutputDataType input2GateWeight;

  OutputDataType input2GateBias;

  OutputDataType gate;

  OutputDataType gateActivation;

  OutputDataType stateActivation;

  OutputDataType cell;

  OutputDataType cellActivation;

  OutputDataType forgetGateError;

  OutputDataType prevError;

  OutputDataType outParameter;

  size_t rhoSize;

  size_t bpttSteps;
}; // class FastLSTM

} // namespace ann
} // namespace mlpack

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

#endif