jsat.linear.distancemetrics

Class TrainableDistanceMetric

java.lang.Object

jsat.linear.distancemetrics.TrainableDistanceMetric

All Implemented Interfaces:

Serializable, Cloneable, DistanceMetric

Direct Known Subclasses:

MahalanobisDistance, NormalizedEuclideanDistance

public abstract class TrainableDistanceMetric
extends Object
implements DistanceMetric

Some Distance Metrics require information that can be learned from the data set. Trainable Distance Metrics support this facility, and algorithms that rely on distance metrics should check if their metric needs training. This is needed priming the distance metric on the whole data set and then performing cross validation would bias the results, as the metric would have been trained on the testing set examples.

Author:

Edward Raff

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
TrainableDistanceMetric()

Method Summary

Modifier and Type	Method and Description
abstract TrainableDistanceMetric	clone()
abstract boolean	needsTraining() Returns true if the metric needs to be trained.
abstract boolean	supportsClassificationTraining() Some metrics might be special purpose, and not trainable for all types of data sets or tasks.
abstract boolean	supportsRegressionTraining() Some metrics might be special purpose, and not trainable for all types of data sets tasks.
abstract void	train(ClassificationDataSet dataSet) Trains this metric on the given classification problem data set
abstract void	train(ClassificationDataSet dataSet, ExecutorService threadpool) Trains this metric on the given classification problem data set
abstract void	train(DataSet dataSet) Trains this metric on the given data set
abstract void	train(DataSet dataSet, ExecutorService threadpool) Trains this metric on the given data set
abstract <V extends Vec> void	train(List<V> dataSet) Trains this metric on the given data set
abstract <V extends Vec> void	train(List<V> dataSet, ExecutorService threadpool) Trains this metric on the given data set
abstract void	train(RegressionDataSet dataSet) Trains this metric on the given regression problem data set
abstract void	train(RegressionDataSet dataSet, ExecutorService threadpool) Trains this metric on the given regression problem data set
static void	trainIfNeeded(DistanceMetric dm, DataSet dataset) Static helper method for training a distance metric only if it is needed.
static void	trainIfNeeded(DistanceMetric dm, DataSet dataset, ExecutorService threadpool) Static helper method for training a distance metric only if it is needed.
static <V extends Vec> void	trainIfNeeded(DistanceMetric dm, List<V> dataset) Static helper method for training a distance metric only if it is needed.
static <V extends Vec> void	trainIfNeeded(DistanceMetric dm, List<V> dataset, ExecutorService threadpool)

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.linear.distancemetrics.DistanceMetric

dist, dist, dist, dist, getAccelerationCache, getAccelerationCache, getQueryInfo, isIndiscemible, isSubadditive, isSymmetric, metricBound, supportsAcceleration, toString

Constructor Detail

TrainableDistanceMetric

public TrainableDistanceMetric()

Method Detail

train

public abstract <V extends Vec> void train(List<V> dataSet)

Trains this metric on the given data set

Type Parameters:

V - the type of vectors in the list

Parameters:

dataSet - the data set to train on

Throws:

UnsupportedOperationException - if the metric can not be trained from unlabeled data

train

public abstract <V extends Vec> void train(List<V> dataSet,
                                           ExecutorService threadpool)

Trains this metric on the given data set

Type Parameters:

V - the type of vectors in the list

Parameters:

dataSet - the data set to train on

threadpool - the source of threads for parallel training

Throws:

UnsupportedOperationException - if the metric can not be trained from unlabeled data

train

public abstract void train(DataSet dataSet)

Trains this metric on the given data set

Parameters:

dataSet - the data set to train on

Throws:

UnsupportedOperationException - if the metric can not be trained from unlabeled data

train

public abstract void train(DataSet dataSet,
                           ExecutorService threadpool)

Trains this metric on the given data set

Parameters:

dataSet - the data set to train on

threadpool - the source of threads for parallel training

Throws:

UnsupportedOperationException - if the metric can not be trained from unlabeled data

train

public abstract void train(ClassificationDataSet dataSet)

Trains this metric on the given classification problem data set

Parameters:

dataSet - the data set to train on

Throws:

UnsupportedOperationException - if the metric can not be trained from classification problems

train

public abstract void train(ClassificationDataSet dataSet,
                           ExecutorService threadpool)

Trains this metric on the given classification problem data set

Parameters:

dataSet - the data set to train on

threadpool - the source of threads for parallel training

Throws:

UnsupportedOperationException - if the metric can not be trained from classification problems

supportsClassificationTraining

public abstract boolean supportsClassificationTraining()

Some metrics might be special purpose, and not trainable for all types of data sets or tasks. This method returns true if this metric supports training for classification problems, and false if it does not.
If a metric can learn from unlabeled data, it must return true for this method.

Returns:

true if this metric supports training for classification problems, and false if it does not

train

public abstract void train(RegressionDataSet dataSet)

Trains this metric on the given regression problem data set

Parameters:

dataSet - the data set to train on

Throws:

UnsupportedOperationException - if the metric can not be trained from regression problems

train

public abstract void train(RegressionDataSet dataSet,
                           ExecutorService threadpool)

Trains this metric on the given regression problem data set

Parameters:

dataSet - the data set to train on

threadpool - the source of threads for parallel training

Throws:

UnsupportedOperationException - if the metric can not be trained from regression problems

supportsRegressionTraining

public abstract boolean supportsRegressionTraining()

Some metrics might be special purpose, and not trainable for all types of data sets tasks. This method returns true if this metric supports training for regression problems, and false if it does not.
If a metric can learn from unlabeled data, it must return true for this method.

Returns:

true if this metric supports training for regression problems, and false if it does not

needsTraining

public abstract boolean needsTraining()

Returns true if the metric needs to be trained. This may be false if the metric allows the parameters to be specified beforehand. If the information was specified before hand, or does not need training, false is returned.

Returns:

true if the metric needs training, false if it does not.

clone

public abstract TrainableDistanceMetric clone()

Specified by:

clone in interface DistanceMetric

Overrides:

clone in class Object

trainIfNeeded

public static void trainIfNeeded(DistanceMetric dm,
                                 DataSet dataset)

Static helper method for training a distance metric only if it is needed. This method can be safely called for any Distance Metric.

Parameters:

dm - the distance metric to train

dataset - the data set to train from

trainIfNeeded

public static void trainIfNeeded(DistanceMetric dm,
                                 DataSet dataset,
                                 ExecutorService threadpool)

Static helper method for training a distance metric only if it is needed. This method can be safely called for any Distance Metric.

Parameters:

dm - the distance metric to train

dataset - the data set to train from

threadpool - the source of threads for parallel training. May be null, in which case trainIfNeeded(jsat.linear.distancemetrics.DistanceMetric, jsat.DataSet) is used instead.

trainIfNeeded

public static <V extends Vec> void trainIfNeeded(DistanceMetric dm,
                                                 List<V> dataset)

Static helper method for training a distance metric only if it is needed. This method can be safely called for any Distance Metric.

Type Parameters:

V - the type of vectors in the list

Parameters:

dm - the distance metric to train

dataset - the data set to train from

trainIfNeeded

public static <V extends Vec> void trainIfNeeded(DistanceMetric dm,
                                                 List<V> dataset,
                                                 ExecutorService threadpool)

Type Parameters:

V - the type of vectors in the list

Parameters:

dm - the distance metric to train

dataset - the data set to train from

threadpool - the source of threads for parallel training. May be null, in which case trainIfNeeded(jsat.linear.distancemetrics.DistanceMetric, java.util.List) is used instead.