HoeffdingNumericSplit< FitnessFunction, ObservationType > Class Template Reference

The HoeffdingNumericSplit class implements the numeric feature splitting strategy alluded to by Domingos and Hulten in the following paper: More...

Public Types

typedef NumericSplitInfo< ObservationType > SplitInfo
 The splitting information type required by the HoeffdingNumericSplit. More...

 

Public Member Functions

 HoeffdingNumericSplit (const size_t numClasses=0, const size_t bins=10, const size_t observationsBeforeBinning=100)
 Create the HoeffdingNumericSplit class, and specify some basic parameters about how the binning should take place. More...

 
 HoeffdingNumericSplit (const size_t numClasses, const HoeffdingNumericSplit &other)
 Create the HoeffdingNumericSplit class, using the parameters from the given other split object. More...

 
size_t Bins () const
 Return the number of bins. More...

 
void EvaluateFitnessFunction (double &bestFitness, double &secondBestFitness) const
 Evaluate the fitness function given what has been calculated so far. More...

 
size_t MajorityClass () const
 Return the majority class. More...

 
double MajorityProbability () const
 Return the probability of the majority class. More...

 
size_t NumChildren () const
 Return the number of children if this node splits on this feature. More...

 
template
<
typename
Archive
>
void serialize (Archive &ar, const unsigned int)
 Serialize the object. More...

 
void Split (arma::Col< size_t > &childMajorities, SplitInfo &splitInfo) const
 Return the majority class of each child to be created, if a split on this dimension was performed. More...

 
void Train (ObservationType value, const size_t label)
 Train the HoeffdingNumericSplit on the given observed value (remember that this object only cares about the information for a single feature, not an entire point). More...

 

Detailed Description


template
<
typename
FitnessFunction
,
typename
ObservationType
=
double
>

class mlpack::tree::HoeffdingNumericSplit< FitnessFunction, ObservationType >

The HoeffdingNumericSplit class implements the numeric feature splitting strategy alluded to by Domingos and Hulten in the following paper:

@inproceedings{domingos2000mining,
title={{Mining High-Speed Data Streams}},
author={Domingos, P. and Hulten, G.},
year={2000},
booktitle={Proceedings of the Sixth ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining (KDD '00)},
pages={71--80}
}

The strategy alluded to is very simple: we discretize the numeric features that we see. But in this case, we don't know how many bins we have, which makes things a little difficult. This class only makes binary splits, and has a maximum number of bins. The binning strategy is simple: the split caches the minimum and maximum value of points seen so far, and when the number of points hits a predefined threshold, the cached minimum-maximum range is equally split into bins, and splitting proceeds in the same way as with the categorical splits. This is a simple and stupid strategy, so don't expect it to be the best possible thing you can do.

Template Parameters
FitnessFunctionFitness function to use for calculating gain.
ObservationTypeType of observations in this dimension.

Definition at line 53 of file hoeffding_numeric_split.hpp.

Member Typedef Documentation

◆ SplitInfo

typedef NumericSplitInfo<ObservationType> SplitInfo

The splitting information type required by the HoeffdingNumericSplit.

Definition at line 57 of file hoeffding_numeric_split.hpp.

Constructor & Destructor Documentation

◆ HoeffdingNumericSplit() [1/2]

HoeffdingNumericSplit ( const size_t  numClasses = 0,
const size_t  bins = 10,
const size_t  observationsBeforeBinning = 100 
)

Create the HoeffdingNumericSplit class, and specify some basic parameters about how the binning should take place.

Parameters
numClassesNumber of classes.
binsNumber of bins.
observationsBeforeBinningNumber of points to see before binning is performed.

◆ HoeffdingNumericSplit() [2/2]

HoeffdingNumericSplit ( const size_t  numClasses,
const HoeffdingNumericSplit< FitnessFunction, ObservationType > &  other 
)

Create the HoeffdingNumericSplit class, using the parameters from the given other split object.

Member Function Documentation

◆ Bins()

size_t Bins ( ) const
inline

Return the number of bins.

Definition at line 120 of file hoeffding_numeric_split.hpp.

References HoeffdingNumericSplit< FitnessFunction, ObservationType >::serialize().

◆ EvaluateFitnessFunction()

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

Evaluate the fitness function given what has been calculated so far.

In this case, if binning has not yet been performed, 0 will be returned (i.e., no gain). Because this split can only split one possible way, secondBestFitness (the fitness function for the second best possible split) will be set to 0.

Parameters
bestFitnessValue of the fitness function for the best possible split.
secondBestFitnessValue of the fitness function for the second best possible split (always 0 for this split).

◆ MajorityClass()

size_t MajorityClass ( ) const

◆ MajorityProbability()

double MajorityProbability ( ) const

Return the probability of the majority class.

Referenced by HoeffdingNumericSplit< FitnessFunction, ObservationType >::NumChildren().

◆ NumChildren()

◆ serialize()

void serialize ( Archive &  ar,
const unsigned  int 
)

◆ Split()

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

Return the majority class of each child to be created, if a split on this dimension was performed.

Also create the split object.

Referenced by HoeffdingNumericSplit< FitnessFunction, ObservationType >::NumChildren().

◆ Train()

void Train ( ObservationType  value,
const size_t  label 
)

Train the HoeffdingNumericSplit on the given observed value (remember that this object only cares about the information for a single feature, not an entire point).

Parameters
valueValue in the dimension that this HoeffdingNumericSplit refers to.
labelLabel of the given point.

The documentation for this class was generated from the following file: