Package edu.gsu.cs.dmlab.sparse.approximation

Class LARS_LassoCoeffVectorApproximator

java.lang.Object

edu.gsu.cs.dmlab.sparse.approximation.LARS_LassoCoeffVectorApproximator

All Implemented Interfaces:

ISparseVectorApproximator

Direct Known Subclasses:

LARS_LassoSparseMatrixApproximator

public class LARS_LassoCoeffVectorApproximator
extends Object
implements ISparseVectorApproximator

Author:

Dustin Kempton, Data Mining Lab, Georgia State University

Constructor Summary

Constructors

Constructor	Description
LARS_LassoCoeffVectorApproximator(double lambda, LassoMode mode)	Constructor

Method Summary

Modifier and Type	Method	Description
void	solve(double[] b, smile.math.matrix.DenseMatrix A, double[] x)	Computes a sparse vector x, such that b = Ax using the LARS algorithm.
void	solve(double[] b, smile.math.matrix.DenseMatrix A, smile.math.matrix.DenseMatrix AtA, double[] x)	Computes a sparse vector x, such that b = Ax using the LARS algorithm.

Methods inherited from class java.lang.Object

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

Constructor Detail

LARS_LassoCoeffVectorApproximator

public LARS_LassoCoeffVectorApproximator(double lambda,
                                         LassoMode mode)

Constructor

Parameters:

lambda - Depending on mode, lambda is used differently. If mode 1, then lambda min amount of correlation between a coefficient in A and signal b for the algorithm to proceed. If mode 2, max L1 norm of the coefficient vector.

mode - What mode to use PENALTY or L1COEFF. See lambda for differences.

Method Detail

solve

public void solve(double[] b,
                  smile.math.matrix.DenseMatrix A,
                  double[] x)
           throws VectorDimensionMismatch,
                  MatrixDimensionMismatch

Computes a sparse vector x, such that b = Ax using the LARS algorithm.

Specified by:

solve in interface ISparseVectorApproximator

Parameters:

b - vector on LHS of equation, which is assumed to have mean 0.

A - Matrix A to use in the solution, which is assumed to have had its columns normalized to have zero mean and unit l2 norm.

x - vector to compute and return

Throws:

VectorDimensionMismatch - If there is some sort of internal error.

MatrixDimensionMismatch

solve

public void solve(double[] b,
                  smile.math.matrix.DenseMatrix A,
                  smile.math.matrix.DenseMatrix AtA,
                  double[] x)
           throws VectorDimensionMismatch,
                  MatrixDimensionMismatch

Computes a sparse vector x, such that b = Ax using the LARS algorithm. Use this method when computing several vectors against the same matrix A to avoid computing AtA every method call.

Specified by:

solve in interface ISparseVectorApproximator

Parameters:

b - vector on LHS of equation, which is assumed to have mean 0.

A - Matrix A to use in the solution, which is assumed to have had its columns normalized to have zero mean and unit l2 norm.

AtA - The Gram Matrix of A used in the solution.

x - vector to compute and return

Throws:

VectorDimensionMismatch - If there is some sort of internal error.

MatrixDimensionMismatch