edu.illinois.cs.cogcomp.lbjava.learn

Class AdaGrad

java.lang.Object

edu.illinois.cs.cogcomp.lbjava.classify.Classifier

edu.illinois.cs.cogcomp.lbjava.learn.Learner

edu.illinois.cs.cogcomp.lbjava.learn.AdaGrad

All Implemented Interfaces:

Serializable, Cloneable

public class AdaGrad
extends Learner

AdaGrad - Adaptive Stochastic Gradient Method AdaGrad alters the update to adapt based on historical information, so that frequent occurring features in the gradients get small learning rates and infrequent features get higher ones. The learner learns slowly from frequent features but "pays attention" to rate but informative features. In practice, this means that infrequently occurring features can be learned effectively along side more frequently occurring features. A good reference for literature is: Duchi, John, Elad Hazan, and Yoram Singer. "Adaptive subgradient methods for online learning and stochastic optimization." The Journal of Machine Learning Research 12 (2011): 2121-2159. http://www.magicbroom.info/Papers/DuchiHaSi10.pdf

Author:

Yiming Jiang (yjiang16@illinois.edu)

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	AdaGrad.Parameters A container for all of AdaGrad's configurable parameters.

Field Summary

Fields

Modifier and Type	Field and Description
static double	defaultLearningRate
static String	defaultLossFunction
protected double	learningRateA
protected String	lossFunctionA

Fields inherited from class edu.illinois.cs.cogcomp.lbjava.learn.Learner

candidates, encoding, extractor, labeler, labelLexicon, lcFilePath, lexFilePath, lexicon, lossFlag, predictions, readLexiconOnDemand

Fields inherited from class edu.illinois.cs.cogcomp.lbjava.classify.Classifier

containingPackage, name

Constructor Summary

Constructors

Constructor and Description
AdaGrad() Constructor The learning rate takes the default value, while the name of the classifier gets the empty string.
AdaGrad(AdaGrad.Parameters p) Constructor Sets all member variables to their associated settings.
AdaGrad(double r) Constructor Sets the learning rate to the specified value, while the name of the classifier gets the empty string.
AdaGrad(String n) Constructor The learning rate takes the default value.
AdaGrad(String n, AdaGrad.Parameters p) Constructor Sets all member variables to their associated settings.
AdaGrad(String n, double r) Constructor Set desired learning rate

Method Summary

Modifier and Type	Method and Description
FeatureVector	classify(int[] exampleFeatures, double[] exampleValues) Simply computes the dot product of the weight vector and the feature vector extracted from the example object.
Feature	featureValue(int[] f, double[] v) Returns the classification of the given example as a single feature instead of a FeatureVector.
double	getConstantLearningRate() Getter - get the constant learning rate
String	getLossFunction() Getter - get loss function
String	getOutputType() Returns a string describing the output feature type of this classifier.
double[]	getWeightVector() Getter - get weight vector
void	learn(int[] exampleFeatures, double[] exampleValues, int[] exampleLabels, double[] labelValues) AdaGrad's Learning Function: Each row of feature vector + label feed in as arguments; Update internal parameters; Note: 1.
double	realValue(int[] exampleFeatures, double[] exampleValues) Simply computes the dot product of the weight vector and the example
ScoreSet	scores(int[] exampleFeatures, double[] exampleValues) Produces a set of scores indicating the degree to which each possible discrete classification value is associated with the given example object.
void	setParameters(AdaGrad.Parameters p) Sets the values of parameters that control the behavior of this learning algorithm.
void	write(PrintStream printStream) Writes the learned function's internal representation as text.

Methods inherited from class edu.illinois.cs.cogcomp.lbjava.learn.Learner

classify, classify, classify, classify, clone, countFeatures, createPrediction, createPrediction, demandLexicon, discreteValue, discreteValue, discreteValue, doneLearning, doneWithRound, emptyClone, featureValue, featureValue, forget, getCurrentLexicon, getExampleArray, getExampleArray, getExtractor, getLabeler, getLabelLexicon, getLexicon, getLexiconDiscardCounts, getLexiconLocation, getModelLocation, getParameters, getPrunedLexiconSize, initialize, learn, learn, learn, learn, read, read, readLabelLexicon, readLearner, readLearner, readLearner, readLearner, readLearner, readLearner, readLexicon, readLexicon, readLexiconOnDemand, readLexiconOnDemand, readModel, readModel, readParameters, realValue, realValue, save, saveLexicon, saveModel, scores, scores, scoresAugmented, setCandidates, setEncoding, setExtractor, setLabeler, setLabelLexicon, setLexicon, setLexiconLocation, setLexiconLocation, setLossFlag, setModelLocation, setModelLocation, setParameters, setReadLexiconOnDemand, unclone, unsetLossFlag, write, write, writeLexicon, writeModel, writeParameters

Methods inherited from class edu.illinois.cs.cogcomp.lbjava.classify.Classifier

allowableValues, classify, discreteValueArray, getCompositeChildren, getInputType, realValueArray, test, toString, valueIndexOf

Methods inherited from class java.lang.Object

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

Field Detail

learningRateA

protected double learningRateA

lossFunctionA

protected String lossFunctionA

defaultLearningRate

public static final double defaultLearningRate

See Also:

Constant Field Values

defaultLossFunction

public static final String defaultLossFunction

See Also:

Constant Field Values

Constructor Detail

AdaGrad

public AdaGrad()

Constructor The learning rate takes the default value, while the name of the classifier gets the empty string.

AdaGrad

public AdaGrad(double r)

Constructor Sets the learning rate to the specified value, while the name of the classifier gets the empty string.

Parameters:

r - The desired learning rate value.

AdaGrad

public AdaGrad(AdaGrad.Parameters p)

Constructor Sets all member variables to their associated settings.

Parameters:

p - The settings of all parameters.

AdaGrad

public AdaGrad(String n)

Constructor The learning rate takes the default value.

Parameters:

n - The name of the classifier.

AdaGrad

public AdaGrad(String n,
               double r)

Constructor Set desired learning rate

Parameters:

n - The name of the classifier.

r - The desired learning rate value.

AdaGrad

public AdaGrad(String n,
               AdaGrad.Parameters p)

Constructor Sets all member variables to their associated settings.

Parameters:

n - The name of the classifier.

p - The settings of all parameters.

Method Detail

setParameters

public void setParameters(AdaGrad.Parameters p)

Sets the values of parameters that control the behavior of this learning algorithm.

Parameters:

p - The parameters.

getWeightVector

public double[] getWeightVector()

Getter - get weight vector

Returns:

weight vector

getLossFunction

public String getLossFunction()

Getter - get loss function

Returns:

"hinge" or "lms"

getConstantLearningRate

public double getConstantLearningRate()

Getter - get the constant learning rate

Returns:

learning rate

learn

public void learn(int[] exampleFeatures,
                  double[] exampleValues,
                  int[] exampleLabels,
                  double[] labelValues)

AdaGrad's Learning Function: Each row of feature vector + label feed in as arguments; Update internal parameters; Note: 1. No bias; No Regularization; are implemented 2. Loss Function used: - Hinge Loss Q((x, y), w) = max(0, 1 - y(w * x)) - Least Mean Square Q((x, y), w) = 1/2 * (y - w * x)^2 3. Notations Explanations: * Feature Vector (exampleValues): feature vector parsed from data set * Label (labelValue): label parsed from data set * Weight Vector (weightVector): weight vector, internal parameter * Gradient (gradientVector): gradient vector, internal parameter for Hinge loss function, g_t = - y_t x_t for LMS loss function, g_t = (w_t * x_t - y_t) x_t where t stands for the t_th iteration * Diagonal Matrix (diagonalVector): diagonal matrix, internal parameter sum of squares of gradients at feature j until time t;

Specified by:

learn in class Learner

Parameters:

exampleFeatures - indices for feature vector x

exampleValues - values for feature vector x

exampleLabels - index for label y

labelValues - value for label y

realValue

public double realValue(int[] exampleFeatures,
                        double[] exampleValues)

Simply computes the dot product of the weight vector and the example

Overrides:

realValue in class Learner

Parameters:

exampleFeatures - The example's array of feature indices.

exampleValues - The example's array of feature values.

Returns:

The computed real value.

featureValue

public Feature featureValue(int[] f,
                            double[] v)

Returns the classification of the given example as a single feature instead of a FeatureVector.

Overrides:

featureValue in class Learner

Parameters:

f - The features array.

v - The values array.

Returns:

The classification of the example as a feature.

classify

public FeatureVector classify(int[] exampleFeatures,
                              double[] exampleValues)

Simply computes the dot product of the weight vector and the feature vector extracted from the example object.

Specified by:

classify in class Learner

Parameters:

exampleFeatures - The example's array of feature indices.

exampleValues - The example's array of feature values.

Returns:

The computed feature (in a vector).

scores

public ScoreSet scores(int[] exampleFeatures,
                       double[] exampleValues)

Produces a set of scores indicating the degree to which each possible discrete classification value is associated with the given example object. Learners that return a real feature or more than one feature may implement this method by simply returning null.

Specified by:

scores in class Learner

Parameters:

exampleFeatures - The example's array of feature indices

exampleValues - The example's array of values

Returns:

A set of scores indicating the degree to which each possible discrete classification value is associated with the given example object.

write

public void write(PrintStream printStream)

Writes the learned function's internal representation as text.

Specified by:

write in class Learner

Parameters:

printStream - The output stream.

getOutputType

public String getOutputType()

Returns a string describing the output feature type of this classifier.

Overrides:

getOutputType in class Classifier

Returns:

"real"