edu.stanford.nlp.math

Class ADMath

java.lang.Object

edu.stanford.nlp.math.ADMath

public class ADMath
extends Object

The class ADMath was created to extend the current calculations of gradient to automatically include a calculation of the hessian vector product with another vector v. It contains all the functions for the DoubleAlgorithmicDifferentiation class. This is used with Stochastic Meta Descent Optimization, but could be extended for use in any application that requires an additional order of differentiation without explicitly creating the code.

Author:

Alex Kleeman

Method Summary

Modifier and Type	Method and Description
static DoubleAD	divide(DoubleAD a, DoubleAD b)
static DoubleAD	divideConst(DoubleAD a, double b)
static DoubleAD	exp(DoubleAD a)
static DoubleAD	log(DoubleAD a)
static DoubleAD	logSum(DoubleAD[] logInputs)
static DoubleAD	logSum(DoubleAD[] logInputs, int fromIndex, int toIndex)
static DoubleAD	minus(DoubleAD a, DoubleAD b)
static DoubleAD	minusConst(DoubleAD a, double b)
static DoubleAD	mult(DoubleAD a, DoubleAD b)
static DoubleAD	multConst(DoubleAD a, double b)
static DoubleAD	plus(DoubleAD a, DoubleAD b)
static DoubleAD	plusConst(DoubleAD a, double b)

Methods inherited from class java.lang.Object

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

Method Detail

mult

public static DoubleAD mult(DoubleAD a,
                            DoubleAD b)

multConst

public static DoubleAD multConst(DoubleAD a,
                                 double b)

divide

public static DoubleAD divide(DoubleAD a,
                              DoubleAD b)

divideConst

public static DoubleAD divideConst(DoubleAD a,
                                   double b)

exp

public static DoubleAD exp(DoubleAD a)

log

public static DoubleAD log(DoubleAD a)

plus

public static DoubleAD plus(DoubleAD a,
                            DoubleAD b)

plusConst

public static DoubleAD plusConst(DoubleAD a,
                                 double b)

minus

public static DoubleAD minus(DoubleAD a,
                             DoubleAD b)

minusConst

public static DoubleAD minusConst(DoubleAD a,
                                  double b)

logSum

public static DoubleAD logSum(DoubleAD[] logInputs)

logSum

public static DoubleAD logSum(DoubleAD[] logInputs,
                              int fromIndex,
                              int toIndex)