jnisvmlight
Class LearnParam

java.lang.Object

  jnisvmlight.LearnParam

public class LearnParam
extends java.lang.Object

Learning parameters as denoted by SVM-light.

Field Summary

java.lang.String	alphafile File to store optimal alphas in.
int	argc The cardinality of the command line parameters.
java.lang.String[]	argv Optionally simulates a simple command shell-like usage and transfers the command line parameters to SVM-light.
long	biased_hyperplane If nonzero, use hyperplane w*x+b=0 otherwise w*x=0.
static int	CLASSIFICATION Trains a classification model.
long	compute_loo If nonzero, computes leave-one-outestimates.
double	eps Regression epsilon (eps=1.0 for classification).
double	epsilon_a Tolerable error on alphas at bounds.
double	epsilon_const Tolerable error on eq-constraint.
double	epsilon_crit Tolerable error for distances used in stopping criterion.
double	epsilon_shrink How much a multiplier should be above zero for shrinking.
long	kernel_cache_size Size of kernel cache in megabytes.
long	maxiter Number of iterations after which the optimizer terminates, if there was no progress in maxdiff.
double	opt_precision Precision of solver, set to e.g.
static int	OPTIMIZATION Trains on general set of constraints.
java.lang.String	predfile File for predicitions on unlabeled examples in transduction.
static int	RANKING Trains a ranking model.
static int	REGRESSION Trains a regression model.
long	remove_inconsistent Exclude examples with alpha at C and retrain.
double	rho Parameter in xi/alpha-estimates and for pruning leave-one-out range [1..2].
long	sharedslack If nonzero, it will use the shared slack variable mode.
long	skip_final_opt_check Do not check KT-Conditions at the end of optimization for examples removed by shrinking.
double	svm_c Upper bound C on alphas.
double	svm_c_factor Increase C by this factor every step.
long	svm_c_steps Do so many steps for finding optimal C.
double	svm_cost Individual upper bounds for each var.
double	svm_costratio Factor to multiply C for positive examples.
double	svm_costratio_unlab
long	svm_iter_to_shrink Iterations h after which an example can be removed by shrinking.
long	svm_maxqpsize Size q of working set.
long	svm_newvarsinqp New variables to enter the working set in each iteration.
double	svm_unlabbound
long	totwords Total amount of features.
double	transduction_posratio Fraction of unlabeled examples to be classified as positives.
long	type Selects between CLASSIFICATION, REGRESSION, RANKING, or OPTIMIZATION mode.
int	verbosity The level of SVM-light debugging infos.
long	xa_depth Parameter in xi/alpha-estimates upper bounding the number of SV the current alpha_t is distributed over.

Constructor Summary

LearnParam()
Initializes the learning parameters with the default SVM-light values.

Method Summary

Methods inherited from class java.lang.Object

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

Field Detail

CLASSIFICATION

public static final int CLASSIFICATION

Trains a classification model.

See Also:

Constant Field Values

OPTIMIZATION

public static final int OPTIMIZATION

Trains on general set of constraints.

See Also:

Constant Field Values

RANKING

public static final int RANKING

Trains a ranking model.

See Also:

Constant Field Values

REGRESSION

public static final int REGRESSION

Trains a regression model.

See Also:

Constant Field Values

alphafile

public java.lang.String alphafile

File to store optimal alphas in. use empty string if alphas should not be output.

argc

public int argc

The cardinality of the command line parameters.

argv

public java.lang.String[] argv

Optionally simulates a simple command shell-like usage and transfers the command line parameters to SVM-light.

biased_hyperplane

public long biased_hyperplane

If nonzero, use hyperplane w*x+b=0 otherwise w*x=0.

compute_loo

public long compute_loo

If nonzero, computes leave-one-outestimates.

eps

public double eps

Regression epsilon (eps=1.0 for classification).

epsilon_a

public double epsilon_a

Tolerable error on alphas at bounds.

epsilon_const

public double epsilon_const

Tolerable error on eq-constraint.

epsilon_crit

public double epsilon_crit

Tolerable error for distances used in stopping criterion.

epsilon_shrink

public double epsilon_shrink

How much a multiplier should be above zero for shrinking.

kernel_cache_size

public long kernel_cache_size

Size of kernel cache in megabytes.

maxiter

public long maxiter

Number of iterations after which the optimizer terminates, if there was no progress in maxdiff.

opt_precision

public double opt_precision

Precision of solver, set to e.g. 1e-21 if you get convergence problems.

predfile

public java.lang.String predfile

File for predicitions on unlabeled examples in transduction.

remove_inconsistent

public long remove_inconsistent

Exclude examples with alpha at C and retrain.

rho

public double rho

Parameter in xi/alpha-estimates and for pruning leave-one-out range [1..2].

sharedslack

public long sharedslack

If nonzero, it will use the shared slack variable mode. In svm_learn_optimization it requires that the slack-id is set for every training example.

skip_final_opt_check

public long skip_final_opt_check

Do not check KT-Conditions at the end of optimization for examples removed by shrinking. WARNING: This might lead to sub-optimal solutions!

svm_c

public double svm_c

Upper bound C on alphas.

svm_c_factor

public double svm_c_factor

Increase C by this factor every step.

svm_c_steps

public long svm_c_steps

Do so many steps for finding optimal C.

svm_cost

public double svm_cost

Individual upper bounds for each var.

svm_costratio

public double svm_costratio

Factor to multiply C for positive examples.

svm_costratio_unlab

public double svm_costratio_unlab

svm_iter_to_shrink

public long svm_iter_to_shrink

Iterations h after which an example can be removed by shrinking.

svm_maxqpsize

public long svm_maxqpsize

Size q of working set.

svm_newvarsinqp

public long svm_newvarsinqp

New variables to enter the working set in each iteration.

svm_unlabbound

public double svm_unlabbound

totwords

public long totwords

Total amount of features.

transduction_posratio

public double transduction_posratio

Fraction of unlabeled examples to be classified as positives.

type

public long type

Selects between CLASSIFICATION, REGRESSION, RANKING, or OPTIMIZATION mode.

verbosity

public int verbosity

The level of SVM-light debugging infos.

xa_depth

public long xa_depth

Parameter in xi/alpha-estimates upper bounding the number of SV the current alpha_t is distributed over.

Constructor Detail

LearnParam

public LearnParam()

Initializes the learning parameters with the default SVM-light values.