org.ddogleg.optimization.impl

Class NumericalJacobianFB

java.lang.Object

org.ddogleg.optimization.impl.NumericalJacobianFB

All Implemented Interfaces:

FunctionNtoMxN

public class NumericalJacobianFB
extends java.lang.Object
implements FunctionNtoMxN

Finite difference numerical jacobian calculation using the forward+backwards equation. Difference equation, f'(x) = (f(x+h)-f(x-h))/(2*h). Scaling is taken in account by h based upon the magnitude of the elements in variable x.

NOTE: If multiple input parameters are modified by the function when a single one is changed numerical derivatives aren't reliable.

Constructor Summary

Constructors

Constructor and Description
NumericalJacobianFB(FunctionNtoM function)
NumericalJacobianFB(FunctionNtoM function, double differenceScale)

Method Summary

Modifier and Type	Method and Description
int	getNumOfInputsN() Number of input parameters and columns in output matrix.
int	getNumOfOutputsM() Number of rows in output matrix.
void	process(double[] input, double[] jacobian) Processes the input vector to output a 2D a matrix.

Methods inherited from class java.lang.Object

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

Constructor Detail

NumericalJacobianFB

public NumericalJacobianFB(FunctionNtoM function,
                           double differenceScale)

NumericalJacobianFB

public NumericalJacobianFB(FunctionNtoM function)

Method Detail

getNumOfInputsN

public int getNumOfInputsN()

Description copied from interface: FunctionNtoMxN

Number of input parameters and columns in output matrix. Typically the parameters you are optimizing.

Specified by:

getNumOfInputsN in interface FunctionNtoMxN

Returns:

Number of input parameters

getNumOfOutputsM

public int getNumOfOutputsM()

Description copied from interface: FunctionNtoMxN

Number of rows in output matrix. Typically the functions that are being optimized.

Specified by:

getNumOfOutputsM in interface FunctionNtoMxN

Returns:

Number of rows in output matrix.

process

public void process(double[] input,
                    double[] jacobian)

Description copied from interface: FunctionNtoMxN

Processes the input vector to output a 2D a matrix. The matrix has a dimension of M rows and N columns and is formatted as a row major 1D-array. EJML can be used to provide a matrix wrapper around the output array: DenseMatrix J = DenseMatrix.wrap(m,n,output);

The user can modify the input parameters here and the optimizer must use those changes.

Specified by:

process in interface FunctionNtoMxN

Parameters:

input - Vector with input parameters.

jacobian - Row major array with M rows and N columns.