org.openimaj.ml.annotation.linear

Class LiblinearAnnotator<OBJECT,ANNOTATION>

java.lang.Object

org.openimaj.ml.annotation.AbstractAnnotator<OBJECT,ANNOTATION>

org.openimaj.ml.annotation.BatchAnnotator<OBJECT,ANNOTATION>

org.openimaj.ml.annotation.linear.LiblinearAnnotator<OBJECT,ANNOTATION>

Type Parameters:

OBJECT - Type of object being annotated

ANNOTATION - Type of annotation

All Implemented Interfaces:

Classifier<ANNOTATION,OBJECT>, Annotator<OBJECT,ANNOTATION>, BatchTrainer<Annotated<OBJECT,ANNOTATION>>

@Reference(type=Article,
           author={"Fan, Rong-En","Chang, Kai-Wei","Hsieh, Cho-Jui","Wang, Xiang-Rui","Lin, Chih-Jen"},
           title="LIBLINEAR: A Library for Large Linear Classification",
           year="2008",
           journal="J. Mach. Learn. Res.",
           pages={"1871","","1874"},
           url="http://dl.acm.org/citation.cfm?id=1390681.1442794",
           month="june",
           publisher="JMLR.org",
           volume="9",
           customData={"date","6/1/2008","issn","1532-4435","numpages","4","acmid","1442794"})
public class LiblinearAnnotator<OBJECT,ANNOTATION>
extends BatchAnnotator<OBJECT,ANNOTATION>

Annotator based on linear classifiers learned using Liblinear (see Linear) or DenseLinear depending on the density of the features. Two modes of operation are available depending on whether the problem is multiclass or multilabel. Binary classification can be achieved with either mode, although multiclass mode is more efficient in this case.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LiblinearAnnotator.Mode The classifier mode; either multiclass or multilabel.

Constructor Summary

Constructors

Constructor and Description
LiblinearAnnotator(FeatureExtractor<? extends FeatureVector,OBJECT> extractor, LiblinearAnnotator.Mode mode, de.bwaldvogel.liblinear.SolverType solver, double C, double eps) Default constructor.
LiblinearAnnotator(FeatureExtractor<? extends FeatureVector,OBJECT> extractor, LiblinearAnnotator.Mode mode, de.bwaldvogel.liblinear.SolverType solver, double C, double eps, double bias, boolean dense) Default constructor.

Method Summary

Methods

Modifier and Type	Method and Description
List<ScoredAnnotation<ANNOTATION>>	annotate(OBJECT object) Generate annotations for the given object.
Set<ANNOTATION>	getAnnotations()
void	train(GroupedDataset<ANNOTATION,ListDataset<OBJECT>,OBJECT> dataset) Train the annotator with the given grouped dataset.
void	train(List<? extends Annotated<OBJECT,ANNOTATION>> data) Train the object with the given data.

Methods inherited from class org.openimaj.ml.annotation.AbstractAnnotator

classify

Methods inherited from class java.lang.Object

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

Constructor Detail

LiblinearAnnotator

public LiblinearAnnotator(FeatureExtractor<? extends FeatureVector,OBJECT> extractor,
                  LiblinearAnnotator.Mode mode,
                  de.bwaldvogel.liblinear.SolverType solver,
                  double C,
                  double eps)

Default constructor. Assumes sparse features.

Parameters:

extractor - the feature extractor

mode - the mode

solver - the liblinear solver

C - the C parameter (usually 1 or larger)

eps - the epsilon value

LiblinearAnnotator

public LiblinearAnnotator(FeatureExtractor<? extends FeatureVector,OBJECT> extractor,
                  LiblinearAnnotator.Mode mode,
                  de.bwaldvogel.liblinear.SolverType solver,
                  double C,
                  double eps,
                  double bias,
                  boolean dense)

Default constructor.

Parameters:

extractor - the feature extractor

mode - the mode

solver - the liblinear solver

C - the C parameter (usually 1 or larger)

eps - the epsilon value

bias - the bias

dense - are the features dense? If so the dense variant of liblinear will be used to drastically reduce memory usage

Method Detail

train

public void train(List<? extends Annotated<OBJECT,ANNOTATION>> data)

Description copied from interface: BatchTrainer

Train the object with the given data. BatchTrainers should reset the underlying model everytime this method is called.

Parameters:

data - the training data

getAnnotations

public Set<ANNOTATION> getAnnotations()

Returns:

a Set of all annotations this Annotator knows about

annotate

public List<ScoredAnnotation<ANNOTATION>> annotate(OBJECT object)

Description copied from interface: Annotator

Generate annotations for the given object.

Parameters:

object - the image

Returns:

generated annotations

train

public void train(GroupedDataset<ANNOTATION,ListDataset<OBJECT>,OBJECT> dataset)

Description copied from class: BatchAnnotator

Train the annotator with the given grouped dataset. Internally, the dataset is converted to a list containing exactly one reference to each object in the dataset with (potentially) multiple annotations.

Overrides:

train in class BatchAnnotator<OBJECT,ANNOTATION>

Parameters:

dataset - the dataset to train on