ai.libs.jaicore.ml.dyadranking.optimizing

Class DyadRankingFeatureTransformNegativeLogLikelihood

java.lang.Object

ai.libs.jaicore.ml.dyadranking.optimizing.DyadRankingFeatureTransformNegativeLogLikelihood

All Implemented Interfaces:

IGradientDescendableFunction, IDyadRankingFeatureTransformPLGradientDescendableFunction

public class DyadRankingFeatureTransformNegativeLogLikelihood
extends java.lang.Object
implements IDyadRankingFeatureTransformPLGradientDescendableFunction

Implements the negative log-likelihood function for the feature transformation Placket-Luce dyad ranker. In particular, this implmentation is the NLL of [1] (we adhere their notation here). This NLL is a convex function, which we can optimize using an IOptimizationAlgorithm, together with the DyadRankingFeatureTransformNegativeLogLikelihoodDerivative. [1] Schäfer, D. & Hüllermeier, Dyad ranking using Plackett–Luce models based on joint feature representations, https://link.springer.com/article/10.1007%2Fs10994-017-5694-9

Constructor Summary

Constructors

Constructor and Description
DyadRankingFeatureTransformNegativeLogLikelihood()

Method Summary

Modifier and Type	Method and Description
double	apply(ai.libs.jaicore.math.linearalgebra.Vector w) Algorithm (18) of [1].
void	initialize(DyadRankingDataset dataset, java.util.Map<IDyadRankingInstance,java.util.Map<Dyad,ai.libs.jaicore.math.linearalgebra.Vector>> featureTransforms) Initializes the function with the given dataset.

Methods inherited from class java.lang.Object

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

Constructor Detail

DyadRankingFeatureTransformNegativeLogLikelihood

public DyadRankingFeatureTransformNegativeLogLikelihood()

Method Detail

initialize

public void initialize(DyadRankingDataset dataset,
                       java.util.Map<IDyadRankingInstance,java.util.Map<Dyad,ai.libs.jaicore.math.linearalgebra.Vector>> featureTransforms)

Description copied from interface: IDyadRankingFeatureTransformPLGradientDescendableFunction

Initializes the function with the given dataset.

Specified by:

initialize in interface IDyadRankingFeatureTransformPLGradientDescendableFunction

Parameters:

dataset - the dataset to use

featureTransforms - the feature precomputed feature transforms

apply

public double apply(ai.libs.jaicore.math.linearalgebra.Vector w)

Algorithm (18) of [1]. We adhere their notations, but, unify the sums.

Specified by:

apply in interface IGradientDescendableFunction

Parameters:

w - the point to which to apply the function

Returns:

the function value at the applied point