batch_norm.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LAYER_BATCHNORM_HPP
#define MLPACK_METHODS_ANN_LAYER_BATCHNORM_HPP

#include <mlpack/prereqs.hpp>

namespace mlpack {
namespace ann  {

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

  BatchNorm(const size_t size,
            const double eps = 1e-8,
            const bool average = true,
            const double momentum = 0.1);

  void Reset();

  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);

  template<typename eT>
  void Gradient(const arma::Mat<eT>& input,
                const arma::Mat<eT>& error,
                arma::Mat<eT>& gradient);

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

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

  OutputDataType const& TrainingMean() const { return runningMean; }
  OutputDataType& TrainingMean() { return runningMean; }

  OutputDataType const& TrainingVariance() const { return runningVariance; }
  OutputDataType& TrainingVariance() { return runningVariance; }

  size_t InputSize() const { return size; }

  double Epsilon() const { return eps; }

  double Momentum() const { return momentum; }

  bool Average() const { return average; }

  size_t WeightSize() const { return 2 * size; }

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

 private:
  size_t size;

  double eps;

  bool average;

  double momentum;

  bool loading;

  OutputDataType gamma;

  OutputDataType beta;

  OutputDataType mean;

  OutputDataType variance;

  OutputDataType weights;

  bool deterministic;

  size_t count;

  double averageFactor;

  OutputDataType runningMean;

  OutputDataType runningVariance;

  OutputDataType gradient;

  OutputDataType delta;

  OutputDataType outputParameter;

  arma::cube normalized;

  arma::cube inputMean;
}; // class BatchNorm

} // namespace ann
} // namespace mlpack

// Include the implementation.
#include "batch_norm_impl.hpp"

#endif