gru.hpp

Go to the documentation of this file.

// Temporarily drop.
#ifndef MLPACK_METHODS_ANN_LAYER_GRU_HPP
#define MLPACK_METHODS_ANN_LAYER_GRU_HPP

#include <list>
#include <limits>

#include <mlpack/prereqs.hpp>

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

namespace mlpack {
namespace ann  {

template <
    typename InputType = arma::mat,
    typename OutputType = arma::mat
>
class GRU : public Layer<InputType, OutputType>
{
 public:
  GRU();

  GRU(const size_t inSize,
      const size_t outSize,
      const size_t rho = std::numeric_limits<size_t>::max());

  void Forward(const InputType& input, OutputType& output);

  void Backward(const InputType& /* input */,
                const OutputType& gy,
                OutputType& 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.
   */
  void Gradient(const InputType& input,
                const OutputType& /* error */,
                OutputType& /* gradient */);

  /*
   * 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);

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

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

  std::vector<Layer<InputType, OutputType>*>& Model() { return network; }

  size_t InSize() const { return inSize; }

  size_t OutSize() const { return outSize; }

  size_t InputShape() const
  {
    return inSize;
  }

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

 private:
  size_t inSize;

  size_t outSize;

  size_t rho;

  size_t batchSize;

  OutputType weights;

  Layer<InputType, OutputType>* input2GateModule;

  Layer<InputType, OutputType>* output2GateModule;

  Layer<InputType, OutputType>* outputHidden2GateModule;

  Layer<InputType, OutputType>* inputGateModule;

  Layer<InputType, OutputType>* hiddenStateModule;

  Layer<InputType, OutputType>* forgetGateModule;

  std::vector<Layer<InputType, OutputType>*> network;

  size_t forwardStep;

  size_t backwardStep;

  size_t gradientStep;

  std::list<OutputType> outParameter;

  OutputType allZeros;

  typename std::list<OutputType>::iterator prevOutput;

  typename std::list<OutputType>::iterator backIterator;

  typename std::list<OutputType>::iterator gradIterator;

  OutputType prevError;

  bool deterministic;

  OutputType delta;

  OutputType gradient;

  OutputType outputParameter;
}; // class GRU

} // namespace ann
} // namespace mlpack

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

#endif