de.uni_stuttgart.isa.liquidsvm

Class SVM

java.lang.Object

de.uni_stuttgart.isa.liquidsvm.SVM

All Implemented Interfaces:

java.io.Serializable

Direct Known Subclasses:

SVM.EX, SVM.LS, SVM.MC, SVM.NPL, SVM.QT, SVM.ROC

public class SVM
extends java.lang.Object
implements java.io.Serializable

liquidSVM SVM base class. After initializing using training data and configuring, at some point training has to be performed and selection of the best hyperparameters. Train and select will be done automagically in most cases. Usually users should use one of the learning scenario subclasses.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	SVM.EX Expectile regression learning scenario.
static class	SVM.LS Least squares regression learning scenario.
static class	SVM.MC Binary and multi-class learning scenario.
static class	SVM.NPL Neyman-Pearson lemma learning scenario.
static class	SVM.QT Quantile regression learning scenario.
static class	SVM.ROC Receiver Operating Characteristic curve learning scenario.

Field Summary

Fields

Modifier and Type	Field and Description
static boolean	ENABLE_LOAD_WITHOUT_LIBRARY
static int	MAX_CPU

Constructor Summary

Constructors

Constructor and Description
SVM() For quick demonstration initialize a least squares SVM using the trees dataset.
SVM(java.lang.String scenario, double[][] data, double[] labels, Config config) Initializes an SVM with training data and optionally configuration options and a scenario.
SVM(java.lang.String scenario, LiquidData.Data data, Config config) Initializes an SVM with training data and optionally configuration options and a scenario.
SVM(java.lang.String scenario, LiquidData data, Config config) Initializes an SVM with training data and optionally configuration options and a scenario and performs testing.

Method Summary

Modifier and Type	Method and Description
static double[][]	calculateDataCover(int NNs, double[][] data) Helper function to calculate Voronoi centers that cover the data set s.t. each cell should not have more than NNs samples.
void	clean()
protected void	finalize()
java.lang.String	getConfig(java.lang.String name) Gets configuration of this SVM
int[]	getCover(int task)
ResultAndErrors	getLastResult() Keeps the last result and errors around after test(double[][], double[], java.lang.String...)ing.
double[][]	getSelectErrs() Returns many information upon select(java.lang.String...)ing.
double[]	getSolutionCoeffs(int task, int cell, int fold)
int[]	getSolutionSVs(int task, int cell, int fold)
double[][]	getTrainErrs() Returns many information upon train(java.lang.String...)ing.
boolean	isSelected() Has the SVM already been selected.
boolean	isTrained() Has the SVM already been trained.
double[]	predict(double[][] test, java.lang.String... argv) Predicts labels for the given test data.
int	readSVM(java.lang.String filename)
double[][]	select(java.lang.String... argv) Selects the best hyperparameter pair according to validation error.
SVM	setConfig(java.lang.String name, double value) Sets configuration directly in the SVM.
SVM	setConfig(java.lang.String name, double[] value) Sets configuration directly in the SVM.
SVM	setConfig(java.lang.String name, int value) Sets configuration directly in the SVM.
SVM	setConfig(java.lang.String name, java.lang.String value) Sets configuration directly in the SVM.
SVM	setConfigAll(Config config) Sets all the values in the config in this SVM.
ResultAndErrors	test(double[][] test, double[] labs, java.lang.String... argv) Predicts labels for the given test data and compares these to the given labs.
ResultAndErrors	test(LiquidData.Data test, java.lang.String... argv) Predicts labels for the given test.x data and compares these to the given test.y.
double[][]	train(java.lang.String... argv) Performs training on the hyperparameter grid.

Methods inherited from class java.lang.Object

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

Field Detail

ENABLE_LOAD_WITHOUT_LIBRARY

public static boolean ENABLE_LOAD_WITHOUT_LIBRARY

MAX_CPU

public static final int MAX_CPU

Constructor Detail

SVM

public SVM()

For quick demonstration initialize a least squares SVM using the trees dataset.

See Also:

LiquidData.trees

SVM

public SVM(java.lang.String scenario,
           double[][] data,
           double[] labels,
           Config config)

Initializes an SVM with training data and optionally configuration options and a scenario. Most users should use rather SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, or SVM.ROC. After the training data and the configuration are set, Config.train(boolean) and Config.select(boolean) determine whether train(String...) and select(String...) should be started. If config==NULL they are performed.

Parameters:

scenario - the learning scenario as in the inner classes or NULL

data - array of training samples features (X)

labels - the training labels (Y)

config - configuration of the SVM. Can be NULL.

See Also:

SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, SVM.ROC

SVM

public SVM(java.lang.String scenario,
           LiquidData.Data data,
           Config config)

Initializes an SVM with training data and optionally configuration options and a scenario. Most users should use rather SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, or SVM.ROC. After the training data and the configuration are set, Config.train(boolean) and Config.select(boolean) determine whether train(String...) and select(String...) should be started. If config==NULL they are performed.

Parameters:

scenario - the learning scenario as in the inner classes or NULL

data - the training data.

config - configuration of the SVM. Can be NULL.

See Also:

SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, SVM.ROC

SVM

public SVM(java.lang.String scenario,
           LiquidData data,
           Config config)

Initializes an SVM with training data and optionally configuration options and a scenario and performs testing. Most users should use rather SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, or SVM.ROC. After the training data and the configuration are set, Config.train(boolean) and Config.select(boolean) determine whether train(String...) and select(String...) should be started. If config==NULL they are performed.

Parameters:

scenario - the learning scenario as in the inner classes or NULL

data - the training and test data.

config - configuration of the SVM. Can be NULL.

See Also:

SVM.LS, SVM.MC, SVM.QT, SVM.EX, SVM.NPL, SVM.ROC

Method Detail

getTrainErrs

public double[][] getTrainErrs()

Returns many information upon train(java.lang.String...)ing. For every point in the hyperparametergrid, every fold, every cell, and every task there is one line.

Returns:

matrix with errors and more information after training.

getSelectErrs

public double[][] getSelectErrs()

Returns many information upon select(java.lang.String...)ing. For every cell, and every task there is one line.

Returns:

matrix with errors and more information after selecting.

getLastResult

public ResultAndErrors getLastResult()

Keeps the last result and errors around after test(double[][], double[], java.lang.String...)ing.

Returns:

the last result and errors.

train

public double[][] train(java.lang.String... argv)

Performs training on the hyperparameter grid. argv can be used by experts in the same way as in the svm-train command line program of liquidSVM interface. Most users should rather use Config.

Parameters:

argv - further command line arguments (for experts)

Returns:

the train errors which can also be retrieved afterwards from getTrainErrs()

select

public double[][] select(java.lang.String... argv)

Selects the best hyperparameter pair according to validation error. argv can be used by experts in the same way as in the svm-select command line program of liquidSVM interface. Most users should rather use Config.

Parameters:

argv - further command line arguments (for experts)

Returns:

the select errors which can also be retrieved afterwards from getSelectErrs()

Throws:

java.lang.IllegalStateException - if the model has not been trained

predict

public double[] predict(double[][] test,
                        java.lang.String... argv)

Predicts labels for the given test data. argv can be used by experts in the same way as in the svm-test command line program of liquidSVM interface. Most users should rather use Config.

Parameters:

test - array of test samples features (X)

argv - further command line arguments (for experts)

Returns:

array of predictions for test data

Throws:

java.lang.IllegalStateException - if the model has neither been trained nor selected

test

public ResultAndErrors test(double[][] test,
                            double[] labs,
                            java.lang.String... argv)

Predicts labels for the given test data and compares these to the given labs. argv can be used by experts in the same way as in the svm-test command line program of liquidSVM interface. Most users should rather use Config.

Parameters:

test - array of test samples features (X)

labs - array of test labels

argv - further command line arguments (for experts)

Returns:

combined predictions and errors

Throws:

java.lang.IllegalStateException - if the model has neither been trained nor selected

test

public ResultAndErrors test(LiquidData.Data test,
                            java.lang.String... argv)

Predicts labels for the given test.x data and compares these to the given test.y. argv can be used by experts in the same way as in the svm-test command line program of liquidSVM interface. Most users should rather use Config.

Parameters:

test - array of test samples

argv - further command line arguments (for experts)

Returns:

combined predictions and errors

Throws:

java.lang.IllegalStateException - if the model has neither been trained nor selected

clean

public void clean()

finalize

protected void finalize()
                 throws java.lang.Throwable

Overrides:

finalize in class java.lang.Object

Throws:

java.lang.Throwable

calculateDataCover

public static double[][] calculateDataCover(int NNs,
                                            double[][] data)

Helper function to calculate Voronoi centers that cover the data set s.t. each cell should not have more than NNs samples.

Parameters:

NNs - biggest size of cells

data - data set to cover

Returns:

array of center vectors

setConfig

public SVM setConfig(java.lang.String name,
                     java.lang.String value)

Sets configuration directly in the SVM. It is better to use Config.

Parameters:

name -

value -

Returns:

this SVM in order to enable method chaining

setConfig

public SVM setConfig(java.lang.String name,
                     int value)

Sets configuration directly in the SVM. It is better to use Config.

Parameters:

name -

value -

Returns:

this SVM in order to enable method chaining

setConfig

public SVM setConfig(java.lang.String name,
                     double value)

Sets configuration directly in the SVM. It is better to use Config.

Parameters:

name -

value -

Returns:

this SVM in order to enable method chaining

setConfig

public SVM setConfig(java.lang.String name,
                     double[] value)

Sets configuration directly in the SVM. It is better to use Config. The values get joined by spaces to form a string.

Parameters:

name -

value -

Returns:

this SVM in order to enable method chaining

setConfigAll

public SVM setConfigAll(Config config)

Sets all the values in the config in this SVM.

Parameters:

config -

Returns:

this SVM in order to enable method chaining

getConfig

public java.lang.String getConfig(java.lang.String name)

Gets configuration of this SVM

Parameters:

name -

Returns:

the configuration value of the key in tihs SVM.

isTrained

public boolean isTrained()

Has the SVM already been trained.

Returns:

whether the SVM already has been trained.

See Also:

train(String...)

isSelected

public boolean isSelected()

Has the SVM already been selected.

Returns:

whether the SVM already has been selected.

See Also:

select(String...)

getCover

public int[] getCover(int task)

getSolutionSVs

public int[] getSolutionSVs(int task,
                            int cell,
                            int fold)

getSolutionCoeffs

public double[] getSolutionCoeffs(int task,
                                  int cell,
                                  int fold)

readSVM

public int readSVM(java.lang.String filename)