KrisLibrary  1.0.0
Public Member Functions | Public Attributes | List of all members
Statistics::StochasticPseudoLASSO Struct Reference

Stochastic estimation of solution to y = A x via a LASSO-like procedure min_x ||b-A x||^2 + alpha*||x||_1. More...

#include <OnlineLASSO.h>

Public Member Functions

 StochasticPseudoLASSO (Real alpha=0.01)
 
void SetPrior (const Vector &coeffs, int strength)
 
void AddPoint (const Vector &data, Real outcome)
 

Public Attributes

Real alpha
 
int numObservations
 
std::vector< Real > weightPolynomial
 
Vector coeffs
 

Detailed Description

Stochastic estimation of solution to y = A x via a LASSO-like procedure min_x ||b-A x||^2 + alpha*||x||_1.

The A matrix is the "data" vector, y is the "outcome" vector. Each update step takes O(n) time, where n is the number of dimensions in x.

The change in coeffs is estimated each step via a LASSO-like step. It is then added to the current estimate via the rule x(m) = x(m-1) + w(m)*deltax(m) where w(m) is a weight equal to 1/P(m), with P(m) being a polynomial.


The documentation for this struct was generated from the following files: