jsat.math

Class OnLineStatistics

java.lang.Object

jsat.math.OnLineStatistics

All Implemented Interfaces:

Serializable, Cloneable

public class OnLineStatistics
extends Object
implements Serializable, Cloneable

This class provides a means of updating summary statistics as each new data point is added. The data points are not stored, and values are updated with an online algorithm.
As such, this class has constant memory usage, regardless of how many values are added. But the results may not be as numerically accurate, and can degrade badly given specific data sequences.

Author:

Edward Raff

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
OnLineStatistics() Creates a new set of statistical counts with no information
OnLineStatistics(double n, double mean, double variance, double skew, double kurt) Creates a new set of statistical counts with these initial values, and can then be updated in an online fashion
OnLineStatistics(OnLineStatistics other) Copy Constructor

Method Summary

Modifier and Type	Method and Description
void	add(double x) Adds a data sample with unit weight to the counts.
void	add(double x, double weight) Adds a data sample the the counts with the provided weight of influence.
void	add(OnLineStatistics B) Adds to the current statistics all the samples that were collected in B.
static OnLineStatistics	add(OnLineStatistics A, OnLineStatistics B) Computes a new set of counts that is the sum of the counts from the given distributions.
OnLineStatistics	clone()
double	getKurtosis()
double	getMax()
double	getMean()
double	getMin()
double	getSkewness()
double	getStandardDeviation()
double	getSumOfWeights() Returns the sum of the weights for all data points added to the statistics.
double	getVarance() Computes the population variance
void	remove(double x, double weight) Effectively removes a sample with the given value and weight from the total.
void	remove(OnLineStatistics B) Removes from this set of statistics the observations that where collected in B. NOTE: removing statistics is not as numerically stable.
static OnLineStatistics	remove(OnLineStatistics A, OnLineStatistics B) Computes a new set of statistics that is the equivalent of having removed all observations in B from A.

Methods inherited from class java.lang.Object

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

Constructor Detail

OnLineStatistics

public OnLineStatistics()

Creates a new set of statistical counts with no information

OnLineStatistics

public OnLineStatistics(double n,
                        double mean,
                        double variance,
                        double skew,
                        double kurt)

Creates a new set of statistical counts with these initial values, and can then be updated in an online fashion

Parameters:

n - the total weight of all data points added. This value must be non negative

mean - the starting mean. If n is zero, this value will be ignored.

variance - the starting variance. If n is zero, this value will be ignored.

skew - the starting skewness. If n is zero, this value will be ignored.

kurt - the starting kurtosis. If n is zero, this value will be ignored.

Throws:

ArithmeticException - if n is a negative number

OnLineStatistics

public OnLineStatistics(OnLineStatistics other)

Copy Constructor

Parameters:

other - the version to make a copy of

Method Detail

add

public void add(double x)

Adds a data sample with unit weight to the counts.

Parameters:

x - the data value to add

add

public void add(double x,
                double weight)

Adds a data sample the the counts with the provided weight of influence.

Parameters:

x - the data value to add

weight - the weight to give the value

Throws:

ArithmeticException - if a negative weight is given

remove

public void remove(double x,
                   double weight)

Effectively removes a sample with the given value and weight from the total. Removing values that have not been added may yield results that have no meaning

NOTE: getSkewness() and getKurtosis() are not currently updated correctly

Parameters:

x - the value of the sample

weight - the weight of the sample

Throws:

ArithmeticException - if a negative weight is given

remove

public static OnLineStatistics remove(OnLineStatistics A,
                                      OnLineStatistics B)

Computes a new set of statistics that is the equivalent of having removed all observations in B from A.
NOTE: removing statistics is not as numerically stable. The values of the 3rd and 4th moments getSkewness() and getKurtosis() will be inaccurate for many inputs. The min and max can not be determined in this setting, and will not be altered.

Parameters:

A - the first set of statistics, which must have a larger value for getSumOfWeights() than B

B - the set of statistics to remove from A.

Returns:

a new set of statistics that is the removal of B from A

remove

public void remove(OnLineStatistics B)

Removes from this set of statistics the observations that where collected in B.
NOTE: removing statistics is not as numerically stable. The values of the 3rd and 4th moments getSkewness() and getKurtosis() will be inaccurate for many inputs. The min and max can not be determined in this setting, and will not be altered.

Parameters:

B - the set of statistics to remove

add

public static OnLineStatistics add(OnLineStatistics A,
                                   OnLineStatistics B)

Computes a new set of counts that is the sum of the counts from the given distributions.

NOTE: Adding two statistics is not as numerically stable. If A and B have values of similar size and scale, the values of the 3rd and 4th moments getSkewness() and getKurtosis() will suffer from catastrophic cancellations, and may not be as accurate.

Parameters:

A - the first set of statistics

B - the second set of statistics

Returns:

a new set of statistics that is the addition of the two.

add

public void add(OnLineStatistics B)

Adds to the current statistics all the samples that were collected in B.
NOTE: Adding two statistics is not as numerically stable. If A and B have values of similar size and scale, the values of the 3rd and 4th moments getSkewness() and getKurtosis() will suffer from catastrophic cancellations, and may not be as accurate.

Parameters:

B - the set of statistics to add to this set

clone

public OnLineStatistics clone()

Overrides:

clone in class Object

getSumOfWeights

public double getSumOfWeights()

Returns the sum of the weights for all data points added to the statistics. If all weights were 1, then this value is the number of data points added.

Returns:

the sum of weights for every point currently contained in the statistics.

getMean

public double getMean()

getVarance

public double getVarance()

Computes the population variance

Returns:

the variance of the data seen

getStandardDeviation

public double getStandardDeviation()

getSkewness

public double getSkewness()

getKurtosis

public double getKurtosis()

getMin

public double getMin()

getMax

public double getMax()