io.improbable.keanu.algorithms.variational

Class GradientOptimizer

java.lang.Object

io.improbable.keanu.algorithms.variational.Optimizer

io.improbable.keanu.algorithms.variational.GradientOptimizer

public class GradientOptimizer
extends Optimizer

Field Summary

Fields

Modifier and Type	Field and Description
static org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer	DEFAULT_OPTIMIZER

Constructor Summary

Constructors

Constructor and Description
GradientOptimizer(BayesianNetwork bayesNet)

Method Summary

Modifier and Type	Method and Description
double	maxAPosteriori()
double	maxAPosteriori(int maxEvaluations)
double	maxAPosteriori(int maxEvaluations, org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer optimizer) This method is here to provide more fine grained control of optimization.
double	maxLikelihood()
double	maxLikelihood(int maxEvaluations)
double	maxLikelihood(int maxEvaluations, org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer optimizer) This method is here to provide more fine grained control of optimization.
void	onFitnessCalculation(java.util.function.BiConsumer<double[],java.lang.Double> fitnessCalculationHandler)
void	onGradientCalculation(java.util.function.BiConsumer<double[],double[]> gradientCalculationHandler)
protected void	warnIfGradientIsFlat(double[] gradient)

Methods inherited from class io.improbable.keanu.algorithms.variational.Optimizer

currentPoint, totalNumLatentDimensions

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_OPTIMIZER

public static final org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer DEFAULT_OPTIMIZER

Constructor Detail

GradientOptimizer

public GradientOptimizer(BayesianNetwork bayesNet)

Method Detail

onGradientCalculation

public void onGradientCalculation(java.util.function.BiConsumer<double[],double[]> gradientCalculationHandler)

onFitnessCalculation

public void onFitnessCalculation(java.util.function.BiConsumer<double[],java.lang.Double> fitnessCalculationHandler)

maxAPosteriori

public double maxAPosteriori(int maxEvaluations,
                             org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer optimizer)

This method is here to provide more fine grained control of optimization.

Parameters:

maxEvaluations - the maximum number of objective function evaluations before throwing an exception indicating convergence failure.

optimizer - apache math optimizer to use for optimization

Returns:

the natural logarithm of the Maximum A Posteriori (MAP)

maxAPosteriori

public double maxAPosteriori(int maxEvaluations)

Parameters:

maxEvaluations - the maximum number of objective function evaluations before throwing an exception indicating convergence failure.

Returns:

the natural logarithm of the Maximum A Posteriori (MAP)

maxAPosteriori

public double maxAPosteriori()

maxLikelihood

public double maxLikelihood(int maxEvaluations,
                            org.apache.commons.math3.optim.nonlinear.scalar.gradient.NonLinearConjugateGradientOptimizer optimizer)

This method is here to provide more fine grained control of optimization.

Parameters:

maxEvaluations - the maximum number of objective function evaluations before throwing an exception indicating convergence failure.

optimizer - apache math optimizer to use for optimization

Returns:

the natural logarithm of the maximum likelihood

maxLikelihood

public double maxLikelihood(int maxEvaluations)

Parameters:

maxEvaluations - the maximum number of objective function evaluations before throwing an exception indicating convergence failure.

Returns:

the natural logarithm of the maximum likelihood

maxLikelihood

public double maxLikelihood()

warnIfGradientIsFlat

protected void warnIfGradientIsFlat(double[] gradient)