reparametrization.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LAYER_REPARAMETRIZATION_HPP
#define MLPACK_METHODS_ANN_LAYER_REPARAMETRIZATION_HPP

#include <mlpack/prereqs.hpp>

#include "layer_types.hpp"
#include "../activation_functions/softplus_function.hpp"

namespace mlpack {
namespace ann  {

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

  Reparametrization(const size_t latentSize,
                    const bool stochastic = true,
                    const bool includeKl = true,
                    const double beta = 1);

  Reparametrization(const Reparametrization& layer);

  Reparametrization(Reparametrization&& layer);

  Reparametrization& operator=(const Reparametrization& layer);

  Reparametrization& operator=(Reparametrization&& layer);

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

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

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

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

  size_t const& OutputSize() const { return latentSize; }
  size_t& OutputSize() { return latentSize; }

  double Loss()
  {
    if (!includeKl)
      return 0;

    return -0.5 * beta * arma::accu(2 * arma::log(stdDev) - arma::pow(stdDev, 2)
        - arma::pow(mean, 2) + 1) / mean.n_cols;
  }

  bool Stochastic() const { return stochastic; }

  bool IncludeKL() const { return includeKl; }

  double Beta() const { return beta; }

  size_t InputShape() const
  {
    return 2 * latentSize;
  }

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

 private:
  size_t latentSize;

  bool stochastic;

  bool includeKl;

  double beta;

  OutputDataType delta;

  OutputDataType gaussianSample;

  OutputDataType mean;

  OutputDataType preStdDev;

  OutputDataType stdDev;

  OutputDataType outputParameter;
}; // class Reparametrization

} // namespace ann
} // namespace mlpack

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

#endif