margin_ranking_loss_impl.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_ANN_LOSS_FUNCTION_MARGIN_IMPL_LOSS_HPP
#define MLPACK_METHODS_ANN_LOSS_FUNCTION_MARGIN_IMPL_LOSS_HPP

// In case it hasn't been included.
#include "margin_ranking_loss.hpp"

namespace mlpack {
namespace ann  {

template<typename InputDataType, typename OutputDataType>
MarginRankingLoss<InputDataType, OutputDataType>::MarginRankingLoss(
    const double margin) : margin(margin)
{
  // Nothing to do here.
}

template<typename InputDataType, typename OutputDataType>
template<typename PredictionType, typename TargetType>
typename PredictionType::elem_type
MarginRankingLoss<InputDataType, OutputDataType>::Forward(
    const PredictionType& prediction,
    const TargetType& target)
{
  const int predictionRows = prediction.n_rows;
  const PredictionType& prediction1 = prediction.rows(0,
      predictionRows / 2 - 1);
  const PredictionType& prediction2 = prediction.rows(predictionRows / 2,
      predictionRows - 1);
  return arma::accu(arma::max(arma::zeros(size(target)),
      -target % (prediction1 - prediction2) + margin)) / target.n_cols;
}

template<typename InputDataType, typename OutputDataType>
template <
    typename PredictionType,
    typename TargetType,
    typename LossType
>
void MarginRankingLoss<InputDataType, OutputDataType>::Backward(
    const PredictionType& prediction,
    const TargetType& target,
    LossType& loss)
{
  const int predictionRows = prediction.n_rows;
  const PredictionType& prediction1 = prediction.rows(0,
      predictionRows / 2 - 1);
  const PredictionType& prediction2 = prediction.rows(predictionRows / 2,
      predictionRows - 1);
  loss = -target % (prediction1 - prediction2) + margin;
  loss.elem(arma::find(loss >= 0)).ones();
  loss.elem(arma::find(loss < 0)).zeros();
  loss = (prediction2 - prediction1) % loss / target.n_cols;
}

template<typename InputDataType, typename OutputDataType>
template<typename Archive>
void MarginRankingLoss<InputDataType, OutputDataType>::serialize(
    Archive& ar,
    const uint32_t /* version */)
{
  ar(CEREAL_NVP(margin));
}

} // namespace ann
} // namespace mlpack

#endif