Package jdistlib.inference.autodiff

Class ReverseModeLogDensity

java.lang.Object

jdistlib.inference.autodiff.ReverseModeLogDensity

All Implemented Interfaces:

DifferentiableLogDensity, GradientProvider, LogDensity

public final class ReverseModeLogDensity
extends Object
implements DifferentiableLogDensity, GradientProvider

Reusable reverse-mode log density suitable for HMC and NUTS.

The primitive tape arena is reset, not reallocated, between evaluations. Instances are deliberately not thread-safe; create one per sampler chain.

Constructor Summary

Constructors

Constructor	Description
ReverseModeLogDensity(int dimension, ReverseDifferentiableFunction function)
ReverseModeLogDensity(int dimension, ReverseDifferentiableFunction function, int initialTapeCapacity)

Method Summary

Modifier and Type	Method	Description
int	dimension()
boolean	hasAnalyticGradient()
double	logDensityAndGradient(double[] state, double[] gradient)	Evaluates the log density and replaces gradient with its gradient.
ReverseTape	tape()

Methods inherited from class java.lang.Object

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

Methods inherited from interface jdistlib.inference.DifferentiableLogDensity

logDensity

Constructor Details

ReverseModeLogDensity

public ReverseModeLogDensity(int dimension,
 ReverseDifferentiableFunction function)

ReverseModeLogDensity

public ReverseModeLogDensity(int dimension,
 ReverseDifferentiableFunction function,
 int initialTapeCapacity)

Method Details

dimension

public int dimension()

tape

public ReverseTape tape()

logDensityAndGradient

public double logDensityAndGradient(double[] state,
 double[] gradient)

Description copied from interface: DifferentiableLogDensity

Evaluates the log density and replaces gradient with its gradient.

Specified by:

logDensityAndGradient in interface DifferentiableLogDensity

hasAnalyticGradient

public boolean hasAnalyticGradient()

Specified by:

hasAnalyticGradient in interface GradientProvider