org.omegahat.Numerics.Optimizers
Class BasicOptimizer

java.lang.Object
  |
  +--org.omegahat.Numerics.Optimizers.BasicOptimizer

All Implemented Interfaces:

OptimizerIterator

Direct Known Subclasses:

NotifyingOptimizer

public class BasicOptimizer

extends java.lang.Object

implements OptimizerIterator

Field Summary

protected ConvergenceChecker	convergenceChecker An optionally specifed object that can determine whether convergence has occurred and that the iteration should terminate.
protected OptimizerAlgorithm	fitter Misnamed field that stores the encapsulation of the objective function and the mechanism for determining the next parameter set, i.e.
protected long	maxNumIterations
protected long	numIterations

Fields inherited from interface org.omegahat.Numerics.Optimizers.OptimizerIterator

CONTINUE, CONVERGED, EXCEPTION, NUMERIC_PROBLEM, TOO_MANY_STEPS

Constructor Summary

BasicOptimizer(OptimizerAlgorithm fitter)

BasicOptimizer(OptimizerAlgorithm fitter, ConvergenceChecker convergence)

Method Summary

long	addToHistory(ParameterSet s)
void	completedIteration(ParameterSet values) Hook method called after each iteration, provided the iteration did not result in convergence.
ConvergenceChecker	convergenceChecker()
ConvergenceChecker	convergenceChecker(ConvergenceChecker c)
ParameterSet	eval()
ParameterSet	eval(ParameterSet initialValues) The basic optimization loop.
OptimizerAlgorithm	fitter()
OptimizerAlgorithm	fitter(OptimizerAlgorithm f)
boolean	handleError(java.lang.Throwable e)
boolean	hasConverged(OptimizerAlgorithm f, ParameterSet values) Test for convergence of the iteration.
long	maxNumIterations()
long	maxNumIterations(long n)
long	numIterations()
long	numIterations(long v)

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.omegahat.Numerics.Optimizers.OptimizerIterator

continueIteration, getState, setState

Field Detail

fitter

protected OptimizerAlgorithm fitter

Misnamed field that stores the encapsulation of the objective function and the mechanism for determining the next parameter set, i.e. the move to a position hopefully closer to the optimum of the objective function.

convergenceChecker

protected ConvergenceChecker convergenceChecker

An optionally specifed object that can determine whether convergence has occurred and that the iteration should terminate. This may be a simple numeric computation, something that stores the history of the parameters and associated "value" of the operand. It can also even solicit input from the user either textually or graphically to determine if an adequate solution has been reached.

numIterations

protected long numIterations

maxNumIterations

protected long maxNumIterations

Constructor Detail

BasicOptimizer

public BasicOptimizer(OptimizerAlgorithm fitter)

BasicOptimizer

public BasicOptimizer(OptimizerAlgorithm fitter,
                      ConvergenceChecker convergence)

Method Detail

hasConverged

public boolean hasConverged(OptimizerAlgorithm f,
                            ParameterSet values)
                     throws java.lang.Exception

Test for convergence of the iteration.

The BasicOptimizer class knows nothing about convergence, except that if someone set the maximum number of iterations and that limit has been exceeded, then it's time to quit.

handleError

public boolean handleError(java.lang.Throwable e)

eval

public ParameterSet eval()

eval

public ParameterSet eval(ParameterSet initialValues)

The basic optimization loop. Calls the algorithm in a loop to compute the next set of parameter values, by invoking the refine method of the fitter.

The object's hasConverged method is called both before and after calling the refine method (if this method does a fair amount of work you might want it to check immediately that it got different parameter values from the previous call).

If the internal convergence test does not break out of the loop, a hook is called, which does nothing much in this class, but notifies listeners in the subclass NotifyingOptimizer. Rather than build in hard-coded tests, the structure of this class and its subclasses allows very general techniques for testing convergence to be added at three different levels:

1. In the algorithm's hasConverged method, for tests that are intrinsic to the numerical technique;
2. By setting a ConvergenceChecker method in this optimizer object itself that makes a convergence test directly;
3. Through an IterationListener object that is notified, makes a test, and throws an InterruptedIterationException, which the eval method will catch (and simply treat as a break out of the iteration).

It is also possible to throw other exceptions from listeners (or from errors in the computations, of course). However, the BasicOptimizer class catches these and ignores them, except for printing an error message and a stack trace.

completedIteration

public void completedIteration(ParameterSet values)

Hook method called after each iteration, provided the iteration did not result in convergence. The interesting version of this method notifies listeners (see the NotifyingOptimizer class).

addToHistory

public long addToHistory(ParameterSet s)

numIterations

public long numIterations()

numIterations

public long numIterations(long v)

maxNumIterations

public long maxNumIterations()

maxNumIterations

public long maxNumIterations(long n)

fitter

public OptimizerAlgorithm fitter()

fitter

public OptimizerAlgorithm fitter(OptimizerAlgorithm f)

convergenceChecker

public ConvergenceChecker convergenceChecker()

convergenceChecker

public ConvergenceChecker convergenceChecker(ConvergenceChecker c)