com.imsl.datamining

Class LogisticRegressionModelObject

java.lang.Object

com.imsl.datamining.LogisticRegressionModelObject

All Implemented Interfaces:

Serializable

public class LogisticRegressionModelObject
extends Object
implements Serializable

Predicts a data set using a previously trained logistic regression model object.

See Also:

Example 1 Example 2, Serialized Form

Constructor Summary

Constructors

Constructor and Description
LogisticRegressionModelObject(LogisticRegression lrm) Constructs a LogisticRegressionModelObject.

Method Summary

Modifier and Type	Method and Description
void	aggregateModels(LogisticRegressionModelObject lrm) Aggregates or combines a different model object to this logistic regression model object.
double[]	getCoefficients() Returns the coefficients.
double[][]	getHessian() Returns the Hessian matrix.
double	getLoglikelihood() Returns the log-likeliood.
int	getNumberOfClasses() Returns the number of classes.
int	getNumberOfCoefficients() Returns the number of coefficients (per class).
PredictiveModel.VariableType	getResponseVariableType() Returns the variable type of the response variable.
double[]	getStandardErrors() Returns the standard errors of the coefficients.
double[]	predict(double[][] testData) Returns the predictions on the test data.
double[]	predict(double[][] testData, double[] testDataWeights) Returns predictions on the given test data based on the given weights.

Methods inherited from class java.lang.Object

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

Constructor Detail

LogisticRegressionModelObject

public LogisticRegressionModelObject(LogisticRegression lrm)

Constructs a LogisticRegressionModelObject.

Parameters:

lrm - a previously trained LogisticRegression model

Method Detail

predict

public double[] predict(double[][] testData)
                 throws PredictiveModel.PredictiveModelException

Returns the predictions on the test data.

Parameters:

testData - a double matrix containing the data. It must be compatible with the original training data (it must have the same variable types and column orders). It is the user's responsibility to ensure the model object and the test data are compatible. Otherwise, errors or unexpected results will occur.

Returns:

a double array containing the predictions

Throws:

PredictiveModel.PredictiveModelException

predict

public double[] predict(double[][] testData,
                        double[] testDataWeights)
                 throws PredictiveModel.PredictiveModelException

Returns predictions on the given test data based on the given weights.

Parameters:

testData - a double matrix containing the test data predictors. It must be compatible with the original training data (it must have the same variable types and column orders). It is the user's responsibility to ensure the model object and the test data are compatible. Otherwise, errors or unexpected results will occur.

testDataWeights - a double array containing the test data weights. For the binomial/multinomial response the weights are typically the frequencies, or number of trials. They must be non-negative.

Returns:

a double array containing predicted outcomes

Throws:

PredictiveModel.PredictiveModelException

aggregateModels

public void aggregateModels(LogisticRegressionModelObject lrm)
                     throws SingularMatrixException

Aggregates or combines a different model object to this logistic regression model object.

It is assumed that the objects have the same structure and are models of the same data problem. The result approximates the model that would result if a LogisticRegression model was trained on all the data. The approximation follows the method presented in Xi, Lin, and Chen (2008).

To illustrate, let \(\beta_1\) and \(\beta_2\) be the MLE’s from separate fits to two different sets of data, and let \(\matrix{H}_1\) and \( \matrix{H}_2\) be the associated Hessian matrices. Then $$\beta = (\matrix{H_1} + \matrix{H_2})^{-1}(\matrix{H_1}\beta_1 + \matrix{H_2}\beta_2) $$ approximates the MLE of the combined data set.

The method replaces the coefficients, standard errors, and Hessian matrix with the aggregated estimates.

Parameters:

lrm - a LogisticRegressionModelObject

Throws:

SingularMatrixException

getCoefficients

public double[] getCoefficients()

Returns the coefficients.

Returns:

a double array containing the coefficient estimates

getStandardErrors

public double[] getStandardErrors()

Returns the standard errors of the coefficients.

Returns:

a double array containing the standard errors

getHessian

public double[][] getHessian()

Returns the Hessian matrix.

Returns:

a double matrix containing the Hessian matrix

getLoglikelihood

public double getLoglikelihood()

Returns the log-likeliood.

Returns:

a double, the log-likelihood of the model

getNumberOfClasses

public int getNumberOfClasses()

Returns the number of classes.

Returns:

an int, the number of classes

getNumberOfCoefficients

public int getNumberOfCoefficients()

Returns the number of coefficients (per class).

Returns:

an int, the number of coefficients

getResponseVariableType

public PredictiveModel.VariableType getResponseVariableType()

Returns the variable type of the response variable.

Returns:

the VariableType of the response variable