genpmreg: Piecewise monotone regression with generalised taut strings

In ftnonpar: Features and Strings for Nonparametric Regression

Description

Applies the generalised taut string method to one-dimensional data.

Usage

genpmreg(y, beta = 0.5, squeezing.factor = 0.5, verbose = FALSE, localsqueezing = TRUE, DYADIC = TRUE, thr.const = 2, extrema.nr = -1, bandwidth = -1, SETTOMEAN = FALSE, method = 2, ...)

Arguments

y	observed values (ordered by value of independent variable)
beta	If method=1 specifies the quantile
squeezing.factor	The amount of decrement applied to the bandwidthes
verbose	logical, if T progress (for each iteration) is illustrated grahically
localsqueezing	logical, if T (default) the bandwidth is changed locally.
DYADIC	If T checks the multiresolution criterion only on dyadic intervals, otherwise all intervals are checked
thr.const	smoothing parameter for the multiresolution criterion (should be approximately 2.5)
extrema.nr	if set to a positive integer an approximation with the specified number of local extreme values is calculated
bandwidth	if set to a positive value the specified bandwidth is used instead of the multiresolution criterion.
SETTOMEAN	logical, if T (default) the value of the taut string approximation at local extreme values is set to the mean or median of the observations on the interval where the extremum is taken.
method	The method used which can be 1 (quantile regression), 2 (usual taut string), 3 (logistic regression) and 4 (Poisson regression)
...	Passed to the plot command if verbose=T

Value

A list with components

y	The approximation of the given data
lambda	Values for lambda used
nmax	Number of local extreme values

Author(s)

Arne Kovac [e-mail]

References

D\"umbgen, L. and Kovac, A. (2006) Extensions of smoothing via taut strings

See Also

pmreg

Examples

data(djdata)
par(mfrow=c(2,2))
plot(djblocks,col="grey")
lines(genpmreg(djblocks,verbose=FALSE,method=2,thr.const=2.5)$y,col="red")
title("Usual taut string method")
ind <- sample(1:length(djblocks),300)
djblocks[ind] <- djblocks[ind]+rnorm(length(ind),0,100)
plot(djblocks,col="grey")
lines(genpmreg(djblocks,verbose=FALSE,method=2)$y,col="red")
title("Usual taut string method with outliers")
plot(djblocks,col="grey")
lines(genpmreg(djblocks,verbose=FALSE,method=1)$y,col="red")
title("Robust taut string method with outliers")
plot(djblocks,col="grey",ylim=c(-10,10))
lines(genpmreg(djblocks,verbose=FALSE,method=1)$y,col="red")
title("Again robust taut string method, different scaling of y-axis")

ftnonpar documentation built on May 2, 2019, 3:04 a.m.