magpie.statistics.utility

Class RobustRegressionUtility

java.lang.Object

magpie.statistics.utility.RobustRegressionUtility

public class RobustRegressionUtility
extends java.lang.Object

Calculates statistics useful to robust regression. Can perform the following operations when provided Dataset that contains both measured and predicted class variables.

• Report the Robust Standard Deviation of Residuals
• Determine which entries are outliers
• Calculate the robust mean and standard deviation in absolute error.

Uses methods outlined in a paper by Motulsky and Brown.

.

Field Summary

Fields

Modifier and Type	Field and Description
protected double	Q Threshold for considering an entry an outlier.
protected double	TestFraction What fraction of the population to test

Constructor Summary

Constructors

Constructor and Description
RobustRegressionUtility()

Method Summary

Modifier and Type	Method and Description
double	getRSDR(double[] errors, double mean, int K) Calculate RSDR if the mean is already known.
double	getRSDR(double[] errors, int K) Calculate the Robust Standard Deviation of Residuals.
boolean[]	isOutlier(Dataset Data) Determines which entries in a dataset qualify as an outlier.
boolean[]	isOutlier(Dataset Data, int K) Determine which entries in a dataset qualify as an outlier, for nonlinear regression purposes.
void	setQ(double q) Define the target False Discovery Rate for outlier selection
void	setTestFraction(double TestFraction) Define the fraction of entries that will be tested as potential outliers.

Methods inherited from class java.lang.Object

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

Field Detail

Q

protected double Q

Threshold for considering an entry an outlier. Expressed as a desired false discovery rate. See paper by Benjamini and Hochberg

TestFraction

protected double TestFraction

What fraction of the population to test

Constructor Detail

RobustRegressionUtility

public RobustRegressionUtility()

Method Detail

setQ

public void setQ(double q)

Define the target False Discovery Rate for outlier selection

Parameters:

q - Desired Q parameter (see Benjamini and Hochberg)

setTestFraction

public void setTestFraction(double TestFraction)

Define the fraction of entries that will be tested as potential outliers.

Parameters:

TestFraction - Fraction of entries to test.

isOutlier

public boolean[] isOutlier(Dataset Data)

Determines which entries in a dataset qualify as an outlier. for nonlinear regression purposes.

Parameters:

Data - Dataset to screen

Returns:

Whether each entry is an outlier

isOutlier

public boolean[] isOutlier(Dataset Data,
                           int K)

Determine which entries in a dataset qualify as an outlier, for nonlinear regression purposes.

Parameters:

Data - Dataset to screen

K - Number of fitting parameters in model

Returns:

Whether each entry is an outlier

getRSDR

public double getRSDR(double[] errors,
                      int K)

Calculate the Robust Standard Deviation of Residuals. Follows the method described in .

NOTE: Their method seems to assume that the mean of residuals is 0, which is likely for most functions. I remove this constraint by subtracting the mean absolute error from the error of each point. This should not change results much. Motulsky and Brown

Parameters:

errors - Errors for entire population (not magnitude of errors).

K - Number of fitted parameters in the model

Returns:

Robust Standard Deviation of Residuals

getRSDR

public double getRSDR(double[] errors,
                      double mean,
                      int K)

Calculate RSDR if the mean is already known. See getRSDR(double[], int) for more details.

Parameters:

errors - Errors for entire population (not magnitude of errors).

K - Number of fitted parameters in the model

mean - Mean error

Returns:

Robust Standard Deviation of Residuals