ai.libs.jaicore.ml.core.optimizing.graddesc

Class GradientDescentOptimizer

java.lang.Object

ai.libs.jaicore.ml.core.optimizing.graddesc.GradientDescentOptimizer

All Implemented Interfaces:

IGradientBasedOptimizer

public class GradientDescentOptimizer
extends java.lang.Object
implements IGradientBasedOptimizer

An optimizer based on the gradient descent method [1]. This optimizer is the naive implementation that calculates the gradient in every step and makes an update into the negative direction of the gradient. This method is known to find the optimum, if the underlying function is convex. At some point in the future, we should probably implement faster methods, like for example http://www.seas.ucla.edu/~vandenbe/236C/lectures/fgrad.pdf [1] Jonathan Barzilai and Jonathan M. Borwein, "Two-point step size gradient methods", in: IMA journal of numerical analysis, 8.1 (1998), pp. 141-148.

Constructor Summary

Constructors

Constructor and Description
GradientDescentOptimizer()
GradientDescentOptimizer(GradientDescentOptimizerConfig config)

Method Summary

Modifier and Type	Method and Description
ai.libs.jaicore.math.linearalgebra.Vector	optimize(IGradientDescendableFunction descendableFunction, IGradientFunction gradient, ai.libs.jaicore.math.linearalgebra.Vector initialGuess) Optimize the given function based on its derivation.

Methods inherited from class java.lang.Object

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

Constructor Detail

GradientDescentOptimizer

public GradientDescentOptimizer(GradientDescentOptimizerConfig config)

Parameters:

config -

GradientDescentOptimizer

public GradientDescentOptimizer()

Method Detail

optimize

public ai.libs.jaicore.math.linearalgebra.Vector optimize(IGradientDescendableFunction descendableFunction,
                                                          IGradientFunction gradient,
                                                          ai.libs.jaicore.math.linearalgebra.Vector initialGuess)

Description copied from interface: IGradientBasedOptimizer

Optimize the given function based on its derivation.

Specified by:

optimize in interface IGradientBasedOptimizer

Parameters:

descendableFunction - the function to optimize

gradient - the first order derivate of the function

initialGuess - the initial guess for the parameters that shall be optimized

Returns:

the optimized vector