hoeffding_numeric_split.hpp

Go to the documentation of this file.

#ifndef MLPACK_METHODS_HOEFFDING_TREES_HOEFFDING_NUMERIC_SPLIT_HPP
#define MLPACK_METHODS_HOEFFDING_TREES_HOEFFDING_NUMERIC_SPLIT_HPP

#include <mlpack/prereqs.hpp>
#include "numeric_split_info.hpp"

namespace mlpack {
namespace tree {

template<typename FitnessFunction,
         typename ObservationType = double>
class HoeffdingNumericSplit
{
 public:
  typedef NumericSplitInfo<ObservationType> SplitInfo;

  HoeffdingNumericSplit(const size_t numClasses,
                        const size_t bins = 10,
                        const size_t observationsBeforeBinning = 100);

  HoeffdingNumericSplit(const size_t numClasses,
                        const HoeffdingNumericSplit& other);

  void Train(ObservationType value, const size_t label);

  void EvaluateFitnessFunction(double& bestFitness, double& secondBestFitness)
      const;

  size_t NumChildren() const { return bins; }

  void Split(arma::Col<size_t>& childMajorities, SplitInfo& splitInfo) const;

  size_t MajorityClass() const;
  double MajorityProbability() const;

  size_t Bins() const { return bins; }

  template<typename Archive>
  void Serialize(Archive& ar, const unsigned int /* version */);

 private:
  arma::Col<ObservationType> observations;
  arma::Col<size_t> labels;

  arma::Col<ObservationType> splitPoints;
  size_t bins;
  size_t observationsBeforeBinning;
  size_t samplesSeen;

  arma::Mat<size_t> sufficientStatistics;
};

template<typename FitnessFunction>
using HoeffdingDoubleNumericSplit = HoeffdingNumericSplit<FitnessFunction,
    double>;

} // namespace tree
} // namespace mlpack

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

#endif