com.rapidminer.operator.learner.functions.kernel.jmysvm.svm
Class SVM

java.lang.Object
  com.rapidminer.operator.learner.functions.kernel.jmysvm.svm.SVM

All Implemented Interfaces:

SVMInterface

Direct Known Subclasses:

SVMpattern, SVMregression

public abstract class SVM
extends java.lang.Object
implements SVMInterface

Abstract base class for all SVMs

Author:

Stefan Rueping, Ingo Mierswa

Field Summary

protected int[]	at_bound
protected double	C
protected double[]	cNeg
protected double	convergence_epsilon
protected double[]	cPos
protected double	descend
protected double	epsilon_neg
protected double	epsilon_pos
protected int	examples_total
protected double	feasible_epsilon
protected double	is_zero
protected double	lambda_eq
protected double	lambda_factor
protected double	lambda_WS
protected int	parameters_working_set_size
protected double[]	primal
protected QuadraticProblem	qp
protected boolean	quadraticLossNeg
protected boolean	quadraticLossPos
protected int	shrink_const
protected double[]	sum
protected double	sum_alpha
protected int	target_count
protected SVMExamples	the_examples
protected Kernel	the_kernel
protected int	to_shrink
protected boolean[]	which_alpha
protected int[]	working_set
protected int	working_set_size

Constructor Summary

SVM()

SVM(Operator paramOperator, Kernel new_kernel, SVMExamples new_examples, ExampleSet rapidMinerExamples, RandomGenerator randomGenerator)
class constructor.

Method Summary

protected void	calculate_working_set() Calculates the working set
protected void	check() check internal variables, for debugging only
protected boolean	convergence() Checks if the optimization converged
protected void	exit_optimizer() exit the optimizer
protected boolean	feasible(int i)
double	getB() Returns the value of b.
double	getC() Gets the complexity constant of the svm.
double[]	getWeights() Return the weights of the features.
double[]	getXiAlphaEstimation(Kernel kernel) Returns a double array of estimated performance values.
protected void	init_optimizer() init the optimizer
protected void	init_working_set() Initialises the working set
void	init(Kernel new_kernel, SVMExamples new_examples) Init the SVM
protected abstract boolean	is_alpha_neg(int i)
protected double	lambda(int i) lagrangion multiplier of variable i
protected void	logln(int level, java.lang.String message) log the output plus newline
protected void	logWarning(java.lang.String message)
protected abstract double	nabla(int i)
protected abstract void	optimize() Calls the optimizer
double	predict(int i)
double	predict(SVMExample sVMExample) predict a single example
void	predict(SVMExamples to_predict) predict values on the testset with model
protected void	print_statistics() print statistics about result
protected void	project_to_constraint() Project variables to constraints
protected void	put_optimizer_values() Stores the optimizer results
protected void	reset_shrinked() reset the shrinked variables
protected void	shrink() shrink the variables
void	train() Train the SVM
protected void	update_working_set() Updates the working set

Methods inherited from class java.lang.Object

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

Field Detail

the_kernel

protected Kernel the_kernel

the_examples

protected SVMExamples the_examples

examples_total

protected int examples_total

target_count

protected int target_count

convergence_epsilon

protected double convergence_epsilon

lambda_factor

protected double lambda_factor

at_bound

protected int[] at_bound

sum

protected double[] sum

which_alpha

protected boolean[] which_alpha

working_set

protected int[] working_set

primal

protected double[] primal

sum_alpha

protected double sum_alpha

lambda_eq

protected double lambda_eq

to_shrink

protected int to_shrink

feasible_epsilon

protected double feasible_epsilon

lambda_WS

protected double lambda_WS

quadraticLossPos

protected boolean quadraticLossPos

quadraticLossNeg

protected boolean quadraticLossNeg

epsilon_pos

protected double epsilon_pos

epsilon_neg

protected double epsilon_neg

working_set_size

protected int working_set_size

parameters_working_set_size

protected int parameters_working_set_size

is_zero

protected double is_zero

shrink_const

protected int shrink_const

C

protected double C

cPos

protected double[] cPos

cNeg

protected double[] cNeg

descend

protected double descend

qp

protected QuadraticProblem qp

Constructor Detail

SVM

public SVM()

SVM

public SVM(Operator paramOperator,
           Kernel new_kernel,
           SVMExamples new_examples,
           ExampleSet rapidMinerExamples,
           RandomGenerator randomGenerator)
    throws UndefinedParameterError

class constructor. Throws an operator exception if a non-optional parameter was not set and has no default value.

Throws:

UndefinedParameterError

Method Detail

init

public void init(Kernel new_kernel,
                 SVMExamples new_examples)

Init the SVM

Specified by:

init in interface SVMInterface

Parameters:

new_kernel - new kernel function.

new_examples - the data container

train

public void train()

Train the SVM

Specified by:

train in interface SVMInterface

print_statistics

protected void print_statistics()

print statistics about result

getWeights

public double[] getWeights()

Return the weights of the features.

Specified by:

getWeights in interface SVMInterface

getB

public double getB()

Returns the value of b.

Specified by:

getB in interface SVMInterface

getC

public double getC()

Gets the complexity constant of the svm.

init_optimizer

protected void init_optimizer()

init the optimizer

exit_optimizer

protected void exit_optimizer()

exit the optimizer

shrink

protected void shrink()

shrink the variables

reset_shrinked

protected void reset_shrinked()

reset the shrinked variables

project_to_constraint

protected void project_to_constraint()

Project variables to constraints

calculate_working_set

protected void calculate_working_set()

Calculates the working set

Throws:

java.lang.Exception - on any error

update_working_set

protected void update_working_set()

Updates the working set

init_working_set

protected void init_working_set()

Initialises the working set

Throws:

java.lang.Exception - on any error

optimize

protected abstract void optimize()

Calls the optimizer

put_optimizer_values

protected void put_optimizer_values()

Stores the optimizer results

convergence

protected boolean convergence()

Checks if the optimization converged

Returns:

boolean true optimzation if converged

nabla

protected abstract double nabla(int i)

lambda

protected double lambda(int i)

lagrangion multiplier of variable i

Parameters:

i - variable index

Returns:

lambda

feasible

protected boolean feasible(int i)

is_alpha_neg

protected abstract boolean is_alpha_neg(int i)

logln

protected void logln(int level,
                     java.lang.String message)

log the output plus newline

Parameters:

level - warning level

message - Message test

logWarning

protected void logWarning(java.lang.String message)

predict

public void predict(SVMExamples to_predict)

predict values on the testset with model

Specified by:

predict in interface SVMInterface

predict

public double predict(int i)

predict

public double predict(SVMExample sVMExample)

predict a single example

Specified by:

predict in interface SVMInterface

check

protected void check()

check internal variables, for debugging only

getXiAlphaEstimation

public double[] getXiAlphaEstimation(Kernel kernel)

Returns a double array of estimated performance values. These are accuracy, precision and recall. Works only for classification SVMs.