jsat.classifiers.svm.extended

Class AMM

java.lang.Object

jsat.classifiers.BaseUpdateableClassifier

jsat.classifiers.svm.extended.OnlineAMM

jsat.classifiers.svm.extended.AMM

All Implemented Interfaces:

Serializable, Cloneable, Classifier, UpdateableClassifier, Parameterized

public class AMM
extends OnlineAMM

This is the batch variant of the Adaptive Multi-Hyperplane Machine (AMM) algorithm. It is related to linear SVMs where instead of having only a single weight vector, it is extended to multi-class problems by giving each class its own weight vector. It is further extended by allowing each class to dynamically add new weight vectors to increase the non-linearity of the solution.
This algorithm works best for problems with a very large number of data points where traditional kernelized SVMs are prohibitively expensive to train due to computational cost.
While the AMM trained in a batch setting can continue to be updated in an online fashion, the accuracy may reduce if done. This is because only the batch variant will reach a local optima.
For this version the BaseUpdateableClassifier.setEpochs(int) method controls the total number of iterations of the learning algorithm. A small value in [5, 20] should be sufficient.
See:

• Wang, Z., Djuric, N., Crammer, K., & Vucetic, S. (2011). Trading representability for scalability Adaptive Multi-Hyperplane Machine for nonlinear Classification. In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD ’11 (p. 24). New York, New York, USA: ACM Press. doi:10.1145/2020408.2020420
• Djuric, N., Lan, L., Vucetic, S., & Wang, Z. (2014). BudgetedSVM: A Toolbox for Scalable SVM Approximations. Journal of Machine Learning Research, 14, 3813–3817. Retrieved from here

Author:

Edward Raff

See Also:

Serialized Form

Field Summary

Fields inherited from class jsat.classifiers.svm.extended.OnlineAMM

c, classBudget, DEFAULT_CLASS_BUDGET, DEFAULT_PRUNE_CONSTANT, DEFAULT_PRUNE_FREQUENCY, DEFAULT_REGULARIZER, k, lambda, nextID, time, weightMatrix

Constructor Summary

Constructors

Constructor and Description
AMM() Creates a new batch AMM learner
AMM(AMM toCopy) Copy constructor
AMM(double lambda) Creates a new batch AMM learner
AMM(double lambda, int classBudget) Creates a new batch AMM learner

Method Summary

Modifier and Type	Method and Description
AMM	clone()
int	getSubEpochs() Returns the number of passes through the data set done on each iteration
void	setSubEpochs(int subEpochs) Each iteration of the batch AMM algorithm requires at least one epoch over the training set.
void	trainC(ClassificationDataSet dataSet) Trains the classifier and constructs a model for classification using the given data set.
void	trainC(ClassificationDataSet dataSet, ExecutorService threadPool) Trains the classifier and constructs a model for classification using the given data set.

Methods inherited from class jsat.classifiers.svm.extended.OnlineAMM

classify, getC, getClassBudget, getLambda, getParameter, getParameters, getPruneFrequency, guessLambda, setC, setClassBudget, setLambda, setPruneFrequency, setUp, supportsWeightedData, update, update

Methods inherited from class jsat.classifiers.BaseUpdateableClassifier

getEpochs, setEpochs, trainEpochs

Methods inherited from class java.lang.Object

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

Constructor Detail

AMM

public AMM()

Creates a new batch AMM learner

AMM

public AMM(double lambda)

Creates a new batch AMM learner

Parameters:

lambda - the regularization value to use

AMM

public AMM(double lambda,
           int classBudget)

Creates a new batch AMM learner

Parameters:

lambda - the regularization value to use

classBudget - the maximum number of weight vectors for each class

AMM

public AMM(AMM toCopy)

Copy constructor

Parameters:

toCopy - the object to copy

Method Detail

setSubEpochs

public void setSubEpochs(int subEpochs)

Each iteration of the batch AMM algorithm requires at least one epoch over the training set. This control how many epochs make up each iteration of training.

Parameters:

subEpochs - the number passes through the training set done on each iteration of training

getSubEpochs

public int getSubEpochs()

Returns the number of passes through the data set done on each iteration

Returns:

the number of passes through the data set done on each iteration

trainC

public void trainC(ClassificationDataSet dataSet,
                   ExecutorService threadPool)

Description copied from interface: Classifier

Trains the classifier and constructs a model for classification using the given data set. If the training method knows how, it will used the threadPool to conduct training in parallel. This method will block until the training has completed.

Specified by:

trainC in interface Classifier

Overrides:

trainC in class BaseUpdateableClassifier

Parameters:

dataSet - the data set to train on

threadPool - the source of threads to use.

trainC

public void trainC(ClassificationDataSet dataSet)

Description copied from interface: Classifier

Trains the classifier and constructs a model for classification using the given data set.

Specified by:

trainC in interface Classifier

Overrides:

trainC in class BaseUpdateableClassifier

Parameters:

dataSet - the data set to train on

clone

public AMM clone()

Specified by:

clone in interface Classifier

Specified by:

clone in interface UpdateableClassifier

Overrides:

clone in class OnlineAMM