softmax_regression.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_SOFTMAX_REGRESSION_SOFTMAX_REGRESSION_HPP
#define MLPACK_METHODS_SOFTMAX_REGRESSION_SOFTMAX_REGRESSION_HPP

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

#include "softmax_regression_function.hpp"

namespace mlpack {
namespace regression {

class SoftmaxRegression
{
 public:
  SoftmaxRegression(const size_t inputSize = 0,
                    const size_t numClasses = 0,
                    const bool fitIntercept = false);
  template<typename OptimizerType = ens::L_BFGS>
  SoftmaxRegression(const arma::mat& data,
                    const arma::Row<size_t>& labels,
                    const size_t numClasses,
                    const double lambda = 0.0001,
                    const bool fitIntercept = false,
                    OptimizerType optimizer = OptimizerType());
  template<typename OptimizerType, typename... CallbackTypes>
  SoftmaxRegression(const arma::mat& data,
                    const arma::Row<size_t>& labels,
                    const size_t numClasses,
                    const double lambda,
                    const bool fitIntercept,
                    OptimizerType optimizer,
                    CallbackTypes&&... callbacks);
  void Classify(const arma::mat& dataset, arma::Row<size_t>& labels) const;
  template<typename VecType>
  size_t Classify(const VecType& point) const;

  void Classify(const arma::mat& dataset,
                arma::Row<size_t>& labels,
                arma::mat& probabilities) const;

  void Classify(const arma::mat& dataset,
                arma::mat& probabilities) const;

  double ComputeAccuracy(const arma::mat& testData,
                         const arma::Row<size_t>& labels) const;
  template<typename OptimizerType = ens::L_BFGS>
  double Train(const arma::mat& data,
               const arma::Row<size_t>& labels,
               const size_t numClasses,
               OptimizerType optimizer = OptimizerType());
  template<typename OptimizerType = ens::L_BFGS, typename... CallbackTypes>
  double Train(const arma::mat& data,
               const arma::Row<size_t>& labels,
               const size_t numClasses,
               OptimizerType optimizer,
               CallbackTypes&&... callbacks);

  size_t& NumClasses() { return numClasses; }
  size_t NumClasses() const { return numClasses; }

  double& Lambda() { return lambda; }
  double Lambda() const { return lambda; }

  bool FitIntercept() const { return fitIntercept; }

  arma::mat& Parameters() { return parameters; }
  const arma::mat& Parameters() const { return parameters; }

  size_t FeatureSize() const
  { return fitIntercept ? parameters.n_cols - 1:
                          parameters.n_cols; }

  template<typename Archive>
  void serialize(Archive& ar, const uint32_t /* version */)
  {
    ar(CEREAL_NVP(parameters));
    ar(CEREAL_NVP(numClasses));
    ar(CEREAL_NVP(lambda));
    ar(CEREAL_NVP(fitIntercept));
  }

 private:
  arma::mat parameters;
  size_t numClasses;
  double lambda;
  bool fitIntercept;
};

} // namespace regression
} // namespace mlpack

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

#endif