de.jstacs.algorithms.optimization
Class DifferentiableFunction

java.lang.Object
  de.jstacs.algorithms.optimization.DifferentiableFunction

All Implemented Interfaces:

Function

Direct Known Subclasses:

LogPrior, NegativeDifferentiableFunction, NumericalDifferentiableFunction, OptimizableFunction

public abstract class DifferentiableFunction
extends Object
implements Function

This class is the framework for any (at least) one time differentiable function f: R^n -> R.

Author:

Jens Keilwagen

Constructor Summary

DifferentiableFunction()

Method Summary

abstract double[]	evaluateGradientOfFunction(double[] x) Evaluates the gradient of function at a certain vector (in mathematical sense) x
protected double[]	findOneDimensionalMin(double[] current, double[] d, double alpha_0, double fAlpha_0, double linEps, double startDistance) This method is used to find an approximation of an onedimensional subfunction.

Methods inherited from class java.lang.Object

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

Methods inherited from interface de.jstacs.algorithms.optimization.Function

evaluateFunction, getDimensionOfScope

Constructor Detail

DifferentiableFunction

public DifferentiableFunction()

Method Detail

evaluateGradientOfFunction

public abstract double[] evaluateGradientOfFunction(double[] x)
                                             throws DimensionException,
                                                    EvaluationException

Evaluates the gradient of function at a certain vector (in mathematical sense) x

Parameters:

x - the current vector

Returns:

the elvaluation of the gradient of a function; has dimension getDimensionOfScope()

Throws:

DimensionException - if dim(x) != n, with f: R^n -> R

EvaluationException - if there was a mistake in evaluating the gradient

See Also:

Function.getDimensionOfScope()

findOneDimensionalMin

protected double[] findOneDimensionalMin(double[] current,
                                         double[] d,
                                         double alpha_0,
                                         double fAlpha_0,
                                         double linEps,
                                         double startDistance)
                                  throws DimensionException,
                                         EvaluationException

This method is used to find an approximation of an onedimensional subfunction. That means it will find an approximation of the minimum starting at point current and search in direction d.

\argmin_{alpha >= 0} f(current + alpha*d)

This method is a standard implementation. You are enabled to overwrite this method to be faster, if you know anything about the problem or if you just want to test other line search methods.

Parameters:

current - the start point

d - the searh direction

alpha_0 - the initial alpha

fAlpha_0 - the initial function value (this value is know in most cases and does not have to be computed again)

linEps - the tolerance for stopping this method

startDistance - the initial distance for bracketing the minimum

Returns:

double[2] res = { alpha*, f(alpha*) }

Throws:

DimensionException

EvaluationException