KrisLibrary
1.0.0
|
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 |
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.