jsat.classifiers.linear

Class LinearTools

java.lang.Object

jsat.classifiers.linear.LinearTools

public class LinearTools
extends java.lang.Object

This class provides static helper methods that may be useful for various linear models.

Method Summary

Modifier and Type	Method and Description
static double	c2Lambda(double C, double N) Many linear classifiers can be phrased in two equivalent forms, that only change the notation for the regularized.
static double	lambda2C(double lambda, double N) Many linear classifiers can be phrased in two equivalent forms, that only change the notation for the regularized.
static double	maxLambdaLogisticL1(ClassificationDataSet cds) If the linear model performs logistic regression regularized by λ ||w||1, this method computes the smallest value of lambda that produces a weight vector of all zeros. Note, that the value returned depends on the data set size.

Methods inherited from class java.lang.Object

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

Method Detail

maxLambdaLogisticL1

public static double maxLambdaLogisticL1(ClassificationDataSet cds)

If the linear model performs logistic regression regularized by λ ||w||₁, this method computes the smallest value of lambda that produces a weight vector of all zeros.

Note, that the value returned depends on the data set size. If being used to initialize the value of λ for cross validation with k-folds, the value (k-1)/k * λ will be closer to the correct value of λ for each CV set.

Parameters:

cds - the data set that the model would be trained from

Returns:

the smallest value of λ that should produce all zeros.

lambda2C

public static double lambda2C(double lambda,
                              double N)

Many linear classifiers can be phrased in two equivalent forms, that only change the notation for the regularized. These forms are:
C ∑_i ℓ(w,x_i) + Ω(w)
and
1/N ∑_i ℓ(w,x_i) + λ Ω(w)
This method converts the regularization parameter λ to the form used as C

Parameters:

lambda - the regularization parameter λ

N - the number of data points in the training set

Returns:

c2Lambda

public static double c2Lambda(double C,
                              double N)

Many linear classifiers can be phrased in two equivalent forms, that only change the notation for the regularized. These forms are:
C ∑_i ℓ(w,x_i) + Ω(w)
and
1/N ∑_i ℓ(w,x_i) + λ Ω(w)
This method converts the regularization parameter C to the form used as λ

Parameters:

C - the regularization parameter C

N - the number of data points in the training set

Returns: