All Packages  Class Hierarchy  This Package  Previous  Next  Index

Class jsvm.Svm

java.lang.Object
    |
    +----jsvm.Svm

public class Svm
extends Object

A Java wrapper for Thorsten Joachims's SVM(light) package. Provides a high-level native interface to invoke C implementations of SVM algorithm.

Version:: $Id$
Author:: Heloise Hse (hwawen@eecs.berkeley.edu)
Author:: Michael Shilman (michaels@eecs.berkeley.edu)

DEBUG_MESSAGES: SVM-Light kernel prints full debug information.
LINEAR_KERNEL: Basic linear kernel type.
NUM_KERNEL_TYPES: The nubmer of kernel types.
NUM_MESSAGE_TYPES: The number of verbosity levels.
POLYNOMIAL_KERNEL: Polynomial kernel: (a*b+1)^d
QUIET_MESSAGES: SVM-Light kernel prints minimal messages.
RADIAL_BASIS_KERNEL: Radial basis function: exp(-gamma ||a-b||^2)
SIGMOID_KERNEL: Sigmoid kernel: tanh(s a*b + c)
SILENT_MESSAGES: SVM-Light kernel prints no messages at all.
USER_KERNEL: User defined kernel from kernel.h
VERBOSE_MESSAGES: SVM-Light kernel prints standard messages.

Svm(): Class constructor.

buildModels(TrainingSet)

Builds a set of models based on the training data in C.

classify(FeatureSet)

Classifies this feature set based on the current model.

classify(TrainingSet)

Classifies test examples, returns the predictions, and calculates the accuracy of the classification.

classify(TrainingType)

Classifies test examples, returns the predictions, and calculates the accuracy of the classification.

classify(Vector)

Classifies test examples and returns the predictions.

getID()

Gets the ID for this Svm object.

getVerbosityLevel()

Gets the current verbosity level.

loadModels(InputStream)

Loads and builds model data structures for classifiers in C.

Reads classifier models from input stream.
Forms a byte array.
Sends the byte array to C.
Reads data from the array and builds data structure in C.

loadTrainingData(InputStream, boolean)

Loads training data.

readLabeledTestCases(InputStream)

Reads labeled test cases from input stream.

readOrigLabeledTestCases(InputStream)

Reads labeled test cases from input stream.

readTestCases(InputStream)

Reads unlabeled test cases from input stream.

removeModels()

Removes existing model data structures in C.

setEpsilon(double)

eps: Allow that error when fitting constraints y [w*x+b] >= 1-eps (default 0.001)

setKernelType(long)

Sets the kernel function type.

setMaxKernelCacheSize(long)

Sets the size of the cache for kernel evaluations (default 1000)

setMaxQPSize(long)

Sets the maximum size of QP-subproblems (default 50)

setPolyD(double)

Sets the parameter d in polynomial kernel.

setRbfG(double)

Sets the parameter gamma in rbf kernel.

setRemoveInconsistent(boolean)

Removes inconsistent training examples and retrain (default 1)

setSigmoidC(double)

Sets the parameter c in sigmoid kernel.

setSigmoidS(double)

Sets the parameter s in sigmoid kernel.

setTradeOffErrorAndMargin(double)

Sets the trade-off between training error and margin (default 1000)

setUserDefinedKernel(String)

Sets the parameter of user defined kernel.

setVerbosityLevel(int)

Verbosity level.

writeModels(OutputStream)

Writes out model text representation to the output stream.

Constructs byte stream from the model data structure in C.
Transfers the byte stream to Java.
Writes it to output stream.

writePredictions(OutputStream, Vector)

Writes the prediction results to an output stream.

SILENT_MESSAGES

public static int SILENT_MESSAGES

SVM-Light kernel prints no messages at all.

QUIET_MESSAGES

public static int QUIET_MESSAGES

SVM-Light kernel prints minimal messages.

VERBOSE_MESSAGES

public static int VERBOSE_MESSAGES

SVM-Light kernel prints standard messages.

DEBUG_MESSAGES

public static int DEBUG_MESSAGES

SVM-Light kernel prints full debug information.

NUM_MESSAGE_TYPES

public static int NUM_MESSAGE_TYPES

The number of verbosity levels.

LINEAR_KERNEL

public static final int LINEAR_KERNEL

Basic linear kernel type.

POLYNOMIAL_KERNEL

public static final int POLYNOMIAL_KERNEL

Polynomial kernel: (a*b+1)^d

RADIAL_BASIS_KERNEL

public static final int RADIAL_BASIS_KERNEL

Radial basis function: exp(-gamma ||a-b||^2)

SIGMOID_KERNEL

public static final int SIGMOID_KERNEL

Sigmoid kernel: tanh(s a*b + c)

USER_KERNEL

public static final int USER_KERNEL

User defined kernel from kernel.h

NUM_KERNEL_TYPES

public static final int NUM_KERNEL_TYPES

The nubmer of kernel types.

Svm

public Svm()

Class constructor.

getID

public long getID()

Gets the ID for this Svm object.

Returns:: the ID for this Svm object.

removeModels

public void removeModels()

Removes existing model data structures in C.

loadTrainingData

public static TrainingSet loadTrainingData(InputStream in,
                                           boolean isNew) throws IOException

Loads training data. Builds data structure on Java side.

Parameters:: in - The inputstream containing the training data.; isNew - Identifies which parser to use. If the data file name has extension, .ntr, set isNew to true. If the extension is .trn, set isNew to false.
Returns:: The training data structure.
Throws: IOException: If an IO error occured.

buildModels

public void buildModels(TrainingSet set) throws SvmException

Builds a set of models based on the training data in C. Calls native methods to allocate memory, and initialize and train on the examples.

Allocates memory for classes.
Initializes classes.
Calls svm_learn to train on the data and build the models.
Removes the training data in C, because it will no longer be used.

Parameters:: set - The data to be trained on.
Throws: SvmException: If models already exist or if there are not training data.

writeModels

public void writeModels(OutputStream out) throws SvmException, IOException

Writes out model text representation to the output stream.

Constructs byte stream from the model data structure in C.
Transfers the byte stream to Java.
Writes it to output stream.

Parameters:: out - The output stream of which the model is written out to.
Throws: SvmException: If models do not exist.
Throws: IOException: If an IO error occured.

loadModels

public void loadModels(InputStream in) throws SvmException, IOException

Loads and builds model data structures for classifiers in C.

Reads classifier models from input stream.
Forms a byte array.
Sends the byte array to C.
Reads data from the array and builds data structure in C.

Parameters:: in - The input stream containing model data.
Throws: SvmException: If models already exist in the system.
Throws: IOException: If an IO error occured.
Another alternative is to parse and build models on Java side, and then transfer the data structure over to the native side. However, we want to minimize the amount of data transferring between Java and native code in order to maintain the efficiency of the program. Therefore, we choose to transfer a byte array and perform data processing using native implementation.

classify

public ClassificationSet classify(FeatureSet s) throws SvmException

Classifies this feature set based on the current model. (Note that this implies that there is a current model; the user must have called buildModels() or loadModels() beforehand).

Parameters:: s - The example to be classified.
Returns:: An enumeration of Classification objects, in order of highest confidence to lowest confidence.
Throws: SvmException: If there's no model to do the classification.

classify

public Vector classify(Vector examples) throws SvmException

Classifies test examples and returns the predictions.

Parameters:: examples - A vector of FeatureSet items, each of which defines a test case.
Returns:: A vector of ClassificationSet elements, each of which contains the confidence values for an example.
Throws: SvmException: If there's no model to do the classification.

classify

public Vector classify(TrainingSet set) throws SvmException

Classifies test examples, returns the predictions, and calculates the accuracy of the classification.

Parameters:: set - Examples from a set of classes.
Returns:: A vector of ClassificationSet elements, each of which contains the confidence values for an example.
Throws: SvmException: If there's no model to do the classification.

classify

public Vector classify(TrainingType type) throws SvmException

Classifies test examples, returns the predictions, and calculates the accuracy of the classification.

Parameters:: type - A set of examples from class type.
Returns:: A vector of ClassificationSet elements, each of which contains the confidence values for an example.
Throws: SvmException: If there's no model to do the classification.

readOrigLabeledTestCases

public static TrainingType readOrigLabeledTestCases(InputStream in) throws IOException

Reads labeled test cases from input stream. Parses and builds data structure.

Reads from .lts files.

Parameters:: in - The input stream containing test examples.
Returns:: A TrainingType data structure containing the test cases with labels for a class.
Throws: IOException: If an IO error occured.

readLabeledTestCases

public static TrainingSet readLabeledTestCases(InputStream in) throws IOException

Reads labeled test cases from input stream. Parses and builds data structure.

Reads from .ntr files.

Parameters:: in - The input stream containing test examples.
Returns:: A TrainingSet data structure containing the test cases with labels for multiple classes.
Throws: IOException: If an IO error occured.

readTestCases

public static Vector readTestCases(InputStream in) throws IOException

Reads unlabeled test cases from input stream. Parses and builds data structure.

Reads from .tst files.

Parameters:: in - The input stream containing test examples.
Returns:: A vector of FeatureSet elements, each of which defines a test case.
Throws: IOException: If an IO error occured.

writePredictions

public static void writePredictions(OutputStream out,
                                    Vector results) throws IOException

Writes the prediction results to an output stream.

Parameters:: out - The output stream to write to.; results - A vector of predictions of type ClassificationSet.
Throws: IOException: If an IO error occured.

setVerbosityLevel

public void setVerbosityLevel(int i)

Verbosity level. Options include SILENT_MESSAGES, QUIET_MESSAGES, VERBOSE_MESSAGES, DEBUG_MESSAGES.

getVerbosityLevel

public int getVerbosityLevel()

Gets the current verbosity level.

Returns:: The current verbosity level.

setRemoveInconsistent

public void setRemoveInconsistent(boolean val)

Removes inconsistent training examples and retrain (default 1)

setMaxQPSize

public void setMaxQPSize(long l)

Sets the maximum size of QP-subproblems (default 50)

setMaxKernelCacheSize

public void setMaxKernelCacheSize(long l)

Sets the size of the cache for kernel evaluations (default 1000)

setTradeOffErrorAndMargin

public void setTradeOffErrorAndMargin(double d)

Sets the trade-off between training error and margin (default 1000)

setEpsilon

public void setEpsilon(double d)

eps: Allow that error when fitting constraints y [w*x+b] >= 1-eps (default 0.001)

setKernelType

public void setKernelType(long l)

Sets the kernel function type. Types include LINEAR_KERNEL, POLYNOMIAL_KERNEL, RADIAL_BASIS_KERNEL, SIGMOID_KERNEL, USER_KERNEL.

setPolyD

public void setPolyD(double l)

Sets the parameter d in polynomial kernel.

setRbfG

public void setRbfG(double d)

Sets the parameter gamma in rbf kernel.

setSigmoidS

public void setSigmoidS(double d)

Sets the parameter s in sigmoid kernel.

setSigmoidC

public void setSigmoidC(double d)

Sets the parameter c in sigmoid kernel.

setUserDefinedKernel

public void setUserDefinedKernel(String s)

Sets the parameter of user defined kernel.

All Packages  Class Hierarchy  This Package  Previous  Next  Index