cggd: Fits Continuous Generalized Gradient Descent models

In cggd: Continuous Generalized Gradient Descent

Description

These CGGD models all provide the entire sequence of coefficients and fits, to the least squares fit. This package is based on the lars package.

Usage

cggd(x, y, beta0 = rep(0,2), kmax = 300, 
                      TRR=FALSE, t0 = 1, TRACE=FALSE,
                      alpha1 = 0, alpha2 = 0, w = 1, 
                      tau = 1, tautil = -1, eps = -1, fctr=1e8)

Arguments

x	matrix of predictors
y	response
beta0	initial coefficient values
kmax	the maximum number of steps
TRACE	If TRUE, cggd prints out its progress
TRR	If TRUE builds a Threshold Ridge Regression (TRR) model. If FALSE builds a Functional Gradient Descent Kernel model.
t0	The initial time in the TRR model.
alpha1,alpha2,w	Values used in kernel generation.
tau,tautil,eps	Values used in variable selection.
fctr	Used to tune the search routine. Smaller values tighten the search (more sampling).

Details

CGGD is a general method for performing different types of regressions while continuously varying regularization and other parameters (similar to LARS but broader). It returns the models fit along the continuous paths of parameter modification. Note that y is automatically centered and x is centered and normalized.

Value

A "cggd" object is returned, for which plot and predict methods exist.

Some significant componenets of the object are:

beta.tk	The coefficients of the model at the different steps (k).
tk	The time at the different steps (k).
a.set.tk	The active coefficients of the model at the different steps (k).

Author(s)

Cun-Hui Zhang and Ofer Melnik

References

Cun-Hui Zhang (2007) "Continuous Generalized Gradient Descent" Journal of Computational and Graphical Statistics ; see also http://stat.rutgers.edu/~cunhui/software/CGGD.html.

See Also

plot, and predict methods for cggd, and cv.cggd

Examples

data(housing)
attach(housing)
object <- cggd(x,y)
plot(object)
object2 <- cggd(x,y,TRR=TRUE)
plot(object2)
detach(housing)

cggd documentation built on May 30, 2017, 4:33 a.m.