edu.stanford.nlp.loglinear.learning

Class LogLikelihoodDifferentiableFunction

java.lang.Object

edu.stanford.nlp.loglinear.learning.AbstractDifferentiableFunction<GraphicalModel>

edu.stanford.nlp.loglinear.learning.LogLikelihoodDifferentiableFunction

public class LogLikelihoodDifferentiableFunction
extends AbstractDifferentiableFunction<GraphicalModel>

Created on 8/23/15.

Author:

keenon

Generates (potentially noisy, no promises about exactness) gradients from a batch of examples that were provided to the system.

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	VARIABLE_TRAINING_VALUE

Constructor Summary

Constructors

Constructor and Description
LogLikelihoodDifferentiableFunction()

Method Summary

Modifier and Type	Method and Description
double	getSummaryForInstance(GraphicalModel model, ConcatVector weights, ConcatVector gradient) Gets a summary of the log-likelihood of a singe model at a point

Methods inherited from class java.lang.Object

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

Field Detail

VARIABLE_TRAINING_VALUE

public static final java.lang.String VARIABLE_TRAINING_VALUE

See Also:

Constant Field Values

Constructor Detail

LogLikelihoodDifferentiableFunction

public LogLikelihoodDifferentiableFunction()

Method Detail

getSummaryForInstance

public double getSummaryForInstance(GraphicalModel model,
                                    ConcatVector weights,
                                    ConcatVector gradient)

Gets a summary of the log-likelihood of a singe model at a point

It assumes that the models have observations for training set as metadata in LogLikelihoodDifferentiableFunction.OBSERVATION_FOR_TRAINING. The models can also have observations fixed in CliqueTree.VARIABLE_OBSERVED_VALUE, but these will be considered fixed and will not be learned against.

Specified by:

getSummaryForInstance in class AbstractDifferentiableFunction<GraphicalModel>

Parameters:

model - the model to find the log-likelihood of

weights - the weights to use

gradient - the gradient to use, will be updated by accumulating the gradient from this instance

Returns:

the gradient and value of the function at that point