org.omegahat.Numerics.Optimizers
Class ConvergenceGradient

java.lang.Object
  |
  +--org.omegahat.Numerics.Optimizers.ConvergenceListener
        |
        +--org.omegahat.Numerics.Optimizers.ConvergenceGradient

All Implemented Interfaces:

ConvergenceListenerInt, java.util.EventListener, IterationListener

public class ConvergenceGradient

extends ConvergenceListener

This listener implements the convergence test that the norm of the gradient is less than the tolerance.

To allow for overall scaling, the smaller of the absolute value and the value relative to the parameters is used.

Fields inherited from class org.omegahat.Numerics.Optimizers.ConvergenceListener

relPrecision, tolerance, value

Constructor Summary

ConvergenceGradient()
A standard object of this class, with a tolerance for convergence equal to the relative precision of computations (about 2e-16).

Method Summary

double

setValue(ModelPoint p, OptimizerIterator opt) Set and return the norm of the gradient.

Methods inherited from class org.omegahat.Numerics.Optimizers.ConvergenceListener

continueIteration, getTolerance, getValue, setTolerance, setValue

Methods inherited from class java.lang.Object

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

Constructor Detail

ConvergenceGradient

public ConvergenceGradient()

A standard object of this class, with a tolerance for convergence equal to the relative precision of computations (about 2e-16).

Method Detail

setValue

public double setValue(ModelPoint p,
                       OptimizerIterator opt)

Set and return the norm of the gradient.

The value is the smaller of the absolute norm and the norm relative to the norm of the parameter vector. If the norm is undefined, returns a very large value to force computation to continue.

Overrides:

setValue in class ConvergenceListener