lpsmooth: non-parametric regression

In bda: Binned Data Analysis

non-parametric regression

Description

To fit nonparametric regression model.

Usage

 lpsmooth(y,x, bw, sd.y,lscv=FALSE, adaptive=FALSE,
 	  from, to, gridsize,conf.level=0.95)
 npr(y,x,sd.x,bw,kernel='decon',optimal=FALSE,adaptive=FALSE,
     x0,from, to, gridsize,conf.level=0.95)
 wlpsmooth(y,x,w,s.x,bw,from,to,gridsize,conf.level=0.95)
 bootsmooth(y,x,type="relative",iter=100,conf.level=0.95)

Arguments

y, x	Two numerical vectors.
w	weights
s.x	standard deviation of the measurement error – Laplacian errors are assumed.
x0, from, to, gridsize	'x0' is the grid points where the fitted values will be evaluated. If it is missing, define a fine grid using the start point ("from"), end point ("to") and size ("gridsize").
bw	Smoothing parameter. Numeric or character value is allowed. If missing, adaptive (LSCV) bandwidth selector will be used.
kernel	kernel type: "normal","gauss","nw","decon" (default), "lp","nadaraya-watson"
lscv, adaptive	If lscv = FALSE, use the given bandwidth to fit lpr directly. If lscv = TRUE and adaptive = FALSE, compute lscv bandwidth and fit lpr. Initial bandwidth should be given. If lscv = TRUE and adaptive = TURE, compute lscv bandwidth, then compute varying smoothing parameter, then fit lpr. This algorithm could be extremeely slow when the sample size is very large.
optimal	Search for optimal bandwidth if TRUE.
sd.y	Standard deviation of y.
sd.x	Standard deviation of the measurement error x.
conf.level	Confidence level.
iter	Bootstrapping iteration number.
type	"relative" changes or "absolute" changes for effectiveness evaluation.

Value

y	Estimated values of the smooth function over a fine grid.
x	grid points where the smoothed function are evaluated.
x0, y0	cleaned data of x and y.
conf.level	confidence level of the simultaneous confidence bands.
pars	estimate parameters including smoothing bandwidth, and parameters for the tube formula.
ucb, lcb	upper and lower confidence bands.
call	function called

Examples

 x <- rnorm(100,34.5,1.5)
 e <- rnorm(100,0,2)
 y <- (x-32)^2 + e
 out <- lpsmooth(y,x)
 out
 plot(out, type='l')
 x0 <- seq(min(x),max(x),length=100)
 y0 <- (x0-32)^2
 lines(x0, y0, col=2)
 points(x, y, pch="*", col=4)


 

bda documentation built on Sept. 11, 2024, 9:01 p.m.