net.sourceforge.nite.datainspection.calc
Class CoincidenceMatrix

java.lang.Object
  net.sourceforge.nite.datainspection.calc.CoincidenceMatrix

public class CoincidenceMatrix
extends java.lang.Object

CoincidenceMatrix has methods for comparing Two Classifications. Both classifications must contain exactly the same items in the same order.
The coincidence matrix accounts for Values (labels)
coincidence_matrix[Val1][Val2] is the number of times that one classification labeled some item Val1 the other Val2
coincidence_matrix is symmetrical along the main-diagonal.
the total over all entry numbers is 2*N, where N is the number of items judged (2*N values are given).
This class contains a method for computing kappa (a type of distance measure between classifications)
This distance uses a distance metric on the type of Values(class labels) assigned to the items(units).
Usually the Boolean Metric is used: distance = 0 iff values are equal otherwise it is 1.
If you think that some labels are more equal than others you may use a weighted kappa, that uses your
own DistanceMetric
kappaKrippendorf - returns the same value as alphaNominal with the standard BooleanMetric.
alphaNominal(DistanceMetric) - according to Krippendorff - requires a DistanceMetric defined on the values of the classification
kappaCohen is computed using the confusion matrix (this kappa may differ from kappa Krippendorff)
The implementation is based on "Computing Krippendorff's Alpha-Reliability"

See Also:

Classification, DistanceMetric

Constructor Summary

CoincidenceMatrix(Classification f, Classification s)

CoincidenceMatrix(ConfusionMatrix conf_m)

CoincidenceMatrix(int size)
creates an empty coincidence matrix of given size with default values the numbers (0,1,...,size-1)

Method Summary

double	alphaNominal(DistanceMetric dist) computes alpha for nominal values using the given distance metric the distance metric should be appropriate for the Values that occur in the Classification for which this CoincidenceMatrix is computed at contruction
double	entry(int row, int col)
double	entry(Value rowValue, Value colValue)
java.util.List	getValues()
double	kappaKrippendorff() returns the same value as alpha when using the standard Boolean Metric.
static void	main(java.lang.String[] args)
int	nrOfItems()
int	numberOfValues()
void	printMatrix(java.io.PrintWriter pw)
void	printMatrix(java.lang.String filename)
void	printValues(java.io.PrintWriter pw) print all Values that occur in first or second on SO
void	setDistanceMetrics(DistanceMetric dist) set the distance metric used
void	setEntry(int row, int col, double cv)
void	setEntry(Value rowValue, Value colValue, double cv) set coinm[rowValue][colValue] to cv
void	setValues(java.util.List vals) set the list of class labels used the order should be the same as the order of Values in the row and columns of the matrix
void	showDistanceMatrix(java.lang.String outFile)
void	showMatrix()
void	showValues() print all Values that occur in first or second on SO

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

CoincidenceMatrix

public CoincidenceMatrix(Classification f,
                         Classification s)

Parameters:

f - the first Classification

s - the second Classification required: f and s are classifications of the same ordered list of items/units classified the constructor computes the coincidence matrix

CoincidenceMatrix

public CoincidenceMatrix(int size)

creates an empty coincidence matrix of given size with default values the numbers (0,1,...,size-1)

CoincidenceMatrix

public CoincidenceMatrix(ConfusionMatrix conf_m)

Parameters:

conf_m - the confusion matrix from which this coincidence matrix is computed (CoinMatrix = ConfMatrix + ConfMatrix^T)

Method Detail

setValues

public void setValues(java.util.List vals)

set the list of class labels used the order should be the same as the order of Values in the row and columns of the matrix

entry

public double entry(int row,
                    int col)

Returns:

coinm[row][col]

entry

public double entry(Value rowValue,
                    Value colValue)

Returns:

coinm[Value r][Value c]

setEntry

public void setEntry(Value rowValue,
                     Value colValue,
                     double cv)

set coinm[rowValue][colValue] to cv

setEntry

public void setEntry(int row,
                     int col,
                     double cv)

getValues

public java.util.List getValues()

showMatrix

public void showMatrix()

printMatrix

public void printMatrix(java.lang.String filename)

printMatrix

public void printMatrix(java.io.PrintWriter pw)

showValues

public void showValues()

print all Values that occur in first or second on SO

printValues

public void printValues(java.io.PrintWriter pw)

print all Values that occur in first or second on SO

numberOfValues

public int numberOfValues()

Returns:

number of class labels used

nrOfItems

public int nrOfItems()

alphaNominal

public double alphaNominal(DistanceMetric dist)

computes alpha for nominal values using the given distance metric
the distance metric should be appropriate for the Values that occur in the Classification for which
this CoincidenceMatrix is computed at contruction

Returns:

alpha = 1.0 - (D_observed / D_chance )

kappaKrippendorff

public double kappaKrippendorff()

returns the same value as alpha when using the standard Boolean Metric. return kappa = (pa-pe)/(1-pe) with pa = 1-Dobserved() and pe = 1-Dchance()

Returns:

kappa according to Krippendorff's alpha method using the standard Boolean Metric

setDistanceMetrics

public void setDistanceMetrics(DistanceMetric dist)

set the distance metric used

showDistanceMatrix

public void showDistanceMatrix(java.lang.String outFile)

main

public static void main(java.lang.String[] args)