jsat.parameters

Class ModelSearch

java.lang.Object

jsat.parameters.ModelSearch

All Implemented Interfaces:

Serializable, Cloneable, Classifier, Regressor

Direct Known Subclasses:

GridSearch, RandomSearch

public abstract class ModelSearch
extends Object
implements Classifier, Regressor

This abstract class provides boilerplate for algorithms that search a model's parameter space to find the parameters that provide the best overall performance.

Author:

Edward Raff

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected Classifier	baseClassifier
protected Regressor	baseRegressor
protected ClassificationScore	classificationTargetScore
protected int	folds The number of CV folds
protected RegressionScore	regressionTargetScore
protected boolean	reuseSameCVFolds If true, create the CV splits once and re-use them for all parameters
protected List<Parameter>	searchParams The list of parameters we will search for, currently only Int and Double params should be used
protected Classifier	trainedClassifier
protected Regressor	trainedRegressor
protected boolean	trainFinalModel If true, trains the final model on the parameters used
protected boolean	trainModelsInParallel If true, parallelism will be obtained by training the models in parallel.

Constructor Summary

Constructors

Constructor and Description
ModelSearch(Classifier baseClassifier, int folds)
ModelSearch(ModelSearch toCopy) Copy constructor
ModelSearch(Regressor baseRegressor, int folds)

Method Summary

Modifier and Type	Method and Description
CategoricalResults	classify(DataPoint data) Performs classification on the given data point.
abstract ModelSearch	clone()
Classifier	getBaseClassifier() Returns the base classifier that was originally passed in when constructing this GridSearch.
Regressor	getBaseRegressor() Returns the base regressor that was originally passed in when constructing this GridSearch.
ClassificationScore	getClassificationTargetScore() Returns the classification score that is trying to be optimized via grid search
protected Parameter	getParameterByName(String name) Finds the parameter object with the given name, or throws an exception if a parameter with the given name does not exist.
RegressionScore	getRegressionTargetScore() Returns the regression score that is trying to be optimized via grid search
Classifier	getTrainedClassifier() Returns the resultant classifier trained on the whole data set after performing parameter tuning.
Regressor	getTrainedRegressor() Returns the resultant regressor trained on the whole data set after performing parameter tuning.
boolean	isReuseSameCVFolds()
boolean	isTrainFinalModel()
boolean	isTrainModelsInParallel()
double	regress(DataPoint data)
void	setClassificationTargetScore(ClassificationScore classifierTargetScore) Sets the score to attempt to optimize when performing grid search on a classification problem.
void	setRegressionTargetScore(RegressionScore regressionTargetScore) Sets the score to attempt to optimize when performing grid search on a regression problem.
void	setReuseSameCVFolds(boolean reuseSameSplit) Sets whether or not one set of CV folds is created and re used for every parameter combination (the default), or if a difference set of CV folds will be used for every parameter combination.
void	setTrainFinalModel(boolean trainFinalModel) If true (the default) the model that was found to be best is trained on the whole data set at the end.
void	setTrainModelsInParallel(boolean trainInParallel) When set to true (the default) parallelism is obtained by training as many models in parallel as possible.
boolean	supportsWeightedData() Indicates whether the model knows how to train using weighted data points.

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.classifiers.Classifier

trainC, trainC

Methods inherited from interface jsat.regression.Regressor

train, train

Field Detail

baseClassifier

protected Classifier baseClassifier

trainedClassifier

protected Classifier trainedClassifier

classificationTargetScore

protected ClassificationScore classificationTargetScore

regressionTargetScore

protected RegressionScore regressionTargetScore

baseRegressor

protected Regressor baseRegressor

trainedRegressor

protected Regressor trainedRegressor

searchParams

protected List<Parameter> searchParams

The list of parameters we will search for, currently only Int and Double params should be used

folds

protected int folds

The number of CV folds

trainModelsInParallel

protected boolean trainModelsInParallel

If true, parallelism will be obtained by training the models in parallel. If false, parallelism is obtained from the model itself.

trainFinalModel

protected boolean trainFinalModel

If true, trains the final model on the parameters used

reuseSameCVFolds

protected boolean reuseSameCVFolds

If true, create the CV splits once and re-use them for all parameters

Constructor Detail

ModelSearch

public ModelSearch(Regressor baseRegressor,
                   int folds)

ModelSearch

public ModelSearch(Classifier baseClassifier,
                   int folds)

ModelSearch

public ModelSearch(ModelSearch toCopy)

Copy constructor

Parameters:

toCopy - the object to copy

Method Detail

setTrainModelsInParallel

public void setTrainModelsInParallel(boolean trainInParallel)

When set to true (the default) parallelism is obtained by training as many models in parallel as possible. If false, parallelsm will be obtained by training the model using the Classifier.trainC(jsat.classifiers.ClassificationDataSet, java.util.concurrent.ExecutorService) and Regressor.train(jsat.regression.RegressionDataSet, java.util.concurrent.ExecutorService) methods.

When a model supports warms starts, parallelism obtained by training the models in parallel is intrinsically reduced, as a model can not be warms started until another model has finished. In the case that one of the parameters is annotated as a warm paramter , that parameter will be the one rained sequential, and for every other parameter combination models will be trained in parallel. If there is no warm parameter, the first parameter added will be used for warm training. If there is only one parameter and warm training is occurring, no parallelism will be obtained.

Parameters:

trainInParallel - true to get parallelism from training many models at the same time, false to get parallelism from getting the model's implicit parallelism.

isTrainModelsInParallel

public boolean isTrainModelsInParallel()

Returns:

true if parallelism is obtained from training many models at the same time, false if parallelism is obtained from using the model's implicit parallelism.

setTrainFinalModel

public void setTrainFinalModel(boolean trainFinalModel)

If true (the default) the model that was found to be best is trained on the whole data set at the end. If false, the final model will not be trained. This means that this Object will not be usable for predictoin. This should only be set if you know you will not be using this model but only want to get the information about which parameter combination is best.

Parameters:

trainFinalModel - true to train the final model after grid search, false to not do that.

isTrainFinalModel

public boolean isTrainFinalModel()

Returns:

true to train the final model after grid search, false to not do that.

setReuseSameCVFolds

public void setReuseSameCVFolds(boolean reuseSameSplit)

Sets whether or not one set of CV folds is created and re used for every parameter combination (the default), or if a difference set of CV folds will be used for every parameter combination.

Parameters:

reuseSameSplit - true if the same split is re-used for every combination, false if a new CV set is used for every parameter combination.

isReuseSameCVFolds

public boolean isReuseSameCVFolds()

Returns:

true if the same split is re-used for every combination, false if a new CV set is used for every parameter combination.

getBaseClassifier

public Classifier getBaseClassifier()

Returns the base classifier that was originally passed in when constructing this GridSearch. If this was not constructed with a classifier, this may return null.

Returns:

the original classifier object given

getTrainedClassifier

public Classifier getTrainedClassifier()

Returns the resultant classifier trained on the whole data set after performing parameter tuning.

Returns:

the trained classifier after a call to Regressor.train(jsat.regression.RegressionDataSet, java.util.concurrent.ExecutorService), or null if it has not been trained.

getBaseRegressor

public Regressor getBaseRegressor()

Returns the base regressor that was originally passed in when constructing this GridSearch. If this was not constructed with a regressor, this may return null.

Returns:

the original regressor object given

getTrainedRegressor

public Regressor getTrainedRegressor()

Returns the resultant regressor trained on the whole data set after performing parameter tuning.

Returns:

the trained regressor after a call to Regressor.train(jsat.regression.RegressionDataSet, java.util.concurrent.ExecutorService), or null if it has not been trained.

setClassificationTargetScore

public void setClassificationTargetScore(ClassificationScore classifierTargetScore)

Sets the score to attempt to optimize when performing grid search on a classification problem.

Parameters:

classifierTargetScore - the score to optimize via grid search

getClassificationTargetScore

public ClassificationScore getClassificationTargetScore()

Returns the classification score that is trying to be optimized via grid search

Returns:

the classification score that is trying to be optimized via grid search

setRegressionTargetScore

public void setRegressionTargetScore(RegressionScore regressionTargetScore)

Sets the score to attempt to optimize when performing grid search on a regression problem.

Parameters:

regressionTargetScore -

getRegressionTargetScore

public RegressionScore getRegressionTargetScore()

Returns the regression score that is trying to be optimized via grid search

Returns:

the regression score that is trying to be optimized via grid search

getParameterByName

protected Parameter getParameterByName(String name)
                                throws IllegalArgumentException

Finds the parameter object with the given name, or throws an exception if a parameter with the given name does not exist.

Parameters:

name - the name to search for

Returns:

the parameter object in question

Throws:

IllegalArgumentException - if the name is not found

classify

public CategoricalResults classify(DataPoint data)

Description copied from interface: Classifier

Performs classification on the given data point.

Specified by:

classify in interface Classifier

Parameters:

data - the data point to classify

Returns:

the results of the classification.

regress

public double regress(DataPoint data)

Specified by:

regress in interface Regressor

supportsWeightedData

public boolean supportsWeightedData()

Description copied from interface: Classifier

Indicates whether the model knows how to train using weighted data points. If it does, the model will train assuming the weights. The values returned by this method may change depending on the parameters set for the model.

Specified by:

supportsWeightedData in interface Classifier

Specified by:

supportsWeightedData in interface Regressor

Returns:

true if the model supports weighted data, false otherwise

clone

public abstract ModelSearch clone()

Specified by:

clone in interface Classifier

Specified by:

clone in interface Regressor

Overrides:

clone in class Object