gini

Documentation

Overview

Package gini contain function to calculating Gini gain.

Gini gain, which is an impurity-based criterion that measures the divergences between the probability distribution of the target attributes' values.

Index

• func FindMaxGain(gains *[]Gini) (MaxGainIdx int)
• func FindMinGiniIndex(ginis *[]Gini) (MinIndexIdx int)
• type Gini
  • func (gini *Gini) ComputeContinu(A *[]float64, T *[]string, C *[]string)
  • func (gini *Gini) ComputeDiscrete(A *[]string, discval *[]string, T *[]string, C *[]string)
  • func (gini *Gini) GetMaxGainValue() float64
  • func (gini *Gini) GetMaxPartGainValue() interface{}
  • func (gini *Gini) GetMinIndexPartValue() interface{}
  • func (gini *Gini) GetMinIndexValue() float64
  • func (gini Gini) String() (s string)

Functions

func FindMaxGain

func FindMaxGain(gains *[]Gini) (MaxGainIdx int)

FindMaxGain find the attribute and value that have the maximum gain. The returned value is index of attribute.

func FindMinGiniIndex

func FindMinGiniIndex(ginis *[]Gini) (MinIndexIdx int)

FindMinGiniIndex return the index of attribute that have the minimum Gini index.

Types

type Gini

type Gini struct {
	// Skip if its true, the gain value would not be searched on this
	// instance.
	Skip bool
	// IsContinue define whether the Gini index came from continuous
	// attribute or not.
	IsContinu bool
	// Value of Gini index for all value in attribute.
	Value float64
	// MaxPartGain contain the index of partition which have the maximum
	// gain.
	MaxPartGain int
	// MaxGainValue contain maximum gain of index.
	MaxGainValue float64
	// MinIndexPart contain the index of partition which have the minimum
	// Gini index.
	MinIndexPart int
	// MinIndexGini contain minimum Gini index value.
	MinIndexValue float64
	// SortedIndex of attribute, sorted by values of attribute. This will
	// be used to reference the unsorted target attribute.
	SortedIndex []int
	// ContinuPart contain list of partition value for continuous attribute.
	ContinuPart []float64
	// DiscretePart contain the possible combination of discrete values.
	DiscretePart tekstus.TableStrings
	// Index contain list of Gini Index for each partition.
	Index []float64
	// Gain contain information gain for each partition.
	Gain []float64
}

Gini contain slice of sorted index, slice of partition values, slice of Gini index, Gini value for all samples.

func (*Gini) ComputeContinu

func (gini *Gini) ComputeContinu(A *[]float64, T *[]string, C *[]string)

ComputeContinu Given an attribute A and the target attribute T which contain N classes in C, compute the information gain of A.

The result of Gini partitions value, Gini Index, and Gini Gain is saved in ContinuPart, Index, and Gain.

func (*Gini) ComputeDiscrete

func (gini *Gini) ComputeDiscrete(A *[]string, discval *[]string, T *[]string,
	C *[]string)

ComputeDiscrete Given an attribute A with discreate value 'discval', and the target attribute T which contain N classes in C, compute the information gain of A.

The result is saved as gain value in MaxGainValue for each partition.

func (*Gini) GetMaxGainValue

func (gini *Gini) GetMaxGainValue() float64

GetMaxGainValue return the value of partition which contain the maximum Gini gain.

func (*Gini) GetMaxPartGainValue

func (gini *Gini) GetMaxPartGainValue() interface{}

GetMaxPartGainValue return the partition that have the maximum Gini gain.

func (*Gini) GetMinIndexPartValue

func (gini *Gini) GetMinIndexPartValue() interface{}

GetMinIndexPartValue return the partition that have the minimum Gini index.

func (*Gini) GetMinIndexValue

func (gini *Gini) GetMinIndexValue() float64

GetMinIndexValue return the minimum Gini index value.

func (Gini) String

func (gini Gini) String() (s string)

String yes, it will print it JSON like format.