jsat.parameters

Class RandomSearch

java.lang.Object

jsat.parameters.ModelSearch

jsat.parameters.RandomSearch

All Implemented Interfaces:

Serializable, Cloneable, Classifier, Regressor

public class RandomSearch
extends ModelSearch

Random Search is a simple method for tuning the parameters of a classification or regression algorithm. Each parameter is given a distribution that represents the values of interest, and trials are done by randomly sampling each parameter from their respective distributions. Compared to GridSearch this method does better when lots of values are to be tested or when 2 or more parameters are to be evaluated.
The model it takes must implement the Parameterized interface. By default, no parameters are selected for optimizations. This is because parameters value ranges are often algorithm specific. As such, the user must specify the parameters and the values to test using the addParameter methods. See : Bergstra, J., & Bengio, Y. (2012). Random Search for Hyper-Parameter Optimization. Journal ofMachine Learning Research, 13, 281–305.

Author:

Edward Raff

See Also:

Serialized Form

Field Summary

Fields inherited from class jsat.parameters.ModelSearch

baseClassifier, baseRegressor, classificationTargetScore, folds, regressionTargetScore, reuseSameCVFolds, searchParams, trainedClassifier, trainedRegressor, trainFinalModel, trainModelsInParallel

Constructor Summary

Constructors

Constructor and Description
RandomSearch(Classifier baseClassifier, int folds) Creates a new GridSearch to tune the specified parameters of a classification model.
RandomSearch(RandomSearch toCopy) Copy constructor
RandomSearch(Regressor baseRegressor, int folds) Creates a new GridSearch to tune the specified parameters of a regression model.

Method Summary

Modifier and Type	Method and Description
void	addParameter(DoubleParameter param, Distribution dist) Adds a new double parameter to be altered for the model being tuned.
void	addParameter(IntParameter param, Distribution dist) Adds a new double parameter to be altered for the model being tuned.
void	addParameter(String name, Distribution dist) Adds a new parameter to be altered for the model being tuned.
int	autoAddParameters(DataSet data) This method will automatically populate the search space with parameters based on which Parameter objects return non-null distributions. Note, using this method with Cross Validation has the potential for over-estimating the accuracy of results if the data set is actually used to for parameter guessing. It is possible for this method to return 0, indicating that no default parameters could be found.
RandomSearch	clone()
int	getTrials()
void	setTrials(int trials) Sets the number of trials or samples that will be taken.
void	train(RegressionDataSet dataSet)
void	train(RegressionDataSet dataSet, ExecutorService threadPool)
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.parameters.ModelSearch

classify, getBaseClassifier, getBaseRegressor, getClassificationTargetScore, getParameterByName, getRegressionTargetScore, getTrainedClassifier, getTrainedRegressor, isReuseSameCVFolds, isTrainFinalModel, isTrainModelsInParallel, regress, setClassificationTargetScore, setRegressionTargetScore, setReuseSameCVFolds, setTrainFinalModel, setTrainModelsInParallel, supportsWeightedData

Methods inherited from class java.lang.Object

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

Constructor Detail

RandomSearch

public RandomSearch(Regressor baseRegressor,
                    int folds)

Creates a new GridSearch to tune the specified parameters of a regression model. The parameters still need to be specified by calling #addParameter(jsat.parameters.DoubleParameter, double[])

Parameters:

baseRegressor - the regressor to tune the parameters of

folds - the number of folds of cross-validation to perform to evaluate each combination of parameters

Throws:

FailedToFitException - if the base regressor does not implement Parameterized

RandomSearch

public RandomSearch(Classifier baseClassifier,
                    int folds)

Creates a new GridSearch to tune the specified parameters of a classification model. The parameters still need to be specified by calling #addParameter(jsat.parameters.DoubleParameter, double[])

Parameters:

baseClassifier - the classifier to tune the parameters of

folds - the number of folds of cross-validation to perform to evaluate each combination of parameters

Throws:

FailedToFitException - if the base classifier does not implement Parameterized

RandomSearch

public RandomSearch(RandomSearch toCopy)

Copy constructor

Parameters:

toCopy - the object to copy

Method Detail

autoAddParameters

public int autoAddParameters(DataSet data)

This method will automatically populate the search space with parameters based on which Parameter objects return non-null distributions.

Note, using this method with Cross Validation has the potential for over-estimating the accuracy of results if the data set is actually used to for parameter guessing.

It is possible for this method to return 0, indicating that no default parameters could be found. The intended interpretation is that there are no parameters that you need to tune to get good performance from the given model. Though there will be cases where the author has simply missed a class.

Parameters:

data - the data set to get parameter estimates from

Returns:

the number of parameters added

setTrials

public void setTrials(int trials)

Sets the number of trials or samples that will be taken. This value is the number of models that will be trained and evaluated for their performance

Parameters:

trials - the number of models to build and evaluate

getTrials

public int getTrials()

Returns:

the number of models that will be built to evaluate

addParameter

public void addParameter(DoubleParameter param,
                         Distribution dist)

Adds a new double parameter to be altered for the model being tuned.

Parameters:

param - the model parameter

initialSearchValues - the distribution to sample from for this parameter

addParameter

public void addParameter(IntParameter param,
                         Distribution dist)

Adds a new double parameter to be altered for the model being tuned.

Parameters:

param - the model parameter

initialSearchValues - the distribution to sample from for this parameter

addParameter

public void addParameter(String name,
                         Distribution dist)

Adds a new parameter to be altered for the model being tuned.

Parameters:

name - the name of the parameter

initialSearchValues - the values to try for the specified parameter

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.

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.

Parameters:

dataSet - the data set to train on

train

public void train(RegressionDataSet dataSet,
                  ExecutorService threadPool)

train

public void train(RegressionDataSet dataSet)

clone

public RandomSearch clone()

Specified by:

clone in interface Classifier

Specified by:

clone in interface Regressor

Specified by:

clone in class ModelSearch