The package rree implements robust relative error estimation. The function rree.generator generates data from positive harmonic distribution, which corresponds to the loss function of the LPRE. We can also generate outliers with arbitrary outlier ratio by changing the argument outlier.ratio. To fit the model, we use the rree function. The function predict is used to make a prediction. For detail, please refer to the help of rree function.
library(rree)## Loading required package: MASS## Loading required package: GeneralizedHyperbolic#generate data
dat <- rree.generator(n=50, p=5, outlier.ratio=0.1) #about last 10% data values are outliers
#fitting
fit <- rree(dat$x, dat$y, gam=0) #ordinary LPRE
fit2 <- rree(dat$x, dat$y, gam=0.2) #robust LPRE with gamma=0.2
fit##
## Call: rree(x = dat$x, y = dat$y, gam = 0)
##
## beta:
## [1] 0.36766 0.25202 0.39204 -0.03252 -0.09134 0.52541
##
## weight:
## [1] 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02
## [15] 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02
## [29] 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02
## [43] 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02fit2##
## Call: rree(x = dat$x, y = dat$y, gam = 0.2)
##
## beta:
## [1] 0.19799 0.38429 0.35980 0.18089 -0.06353 0.25913
##
## weight:
## [1] 0.0200894 0.0247332 0.0198692 0.0193874 0.0248939 0.0249067 0.0193572
## [8] 0.0203329 0.0179628 0.0267462 0.0246585 0.0223991 0.0214285 0.0192457
## [15] 0.0203681 0.0291554 0.0251925 0.0190952 0.0210561 0.0180471 0.0165067
## [22] 0.0173746 0.0246328 0.0272392 0.0199662 0.0140033 0.0210588 0.0344358
## [29] 0.0208942 0.0239129 0.0232463 0.0225034 0.0175451 0.0118743 0.0165469
## [36] 0.0219116 0.0203814 0.0216468 0.0234695 0.0215921 0.0213107 0.0180821
## [43] 0.0221992 0.0209106 0.0226583 0.0237275 0.0034467 0.0054429 0.0019575
## [50] 0.0005974#prediction
predict(fit2, newx=dat$x[1:10,])## [1] 0.8400173 0.3522975 1.9229040 3.2875169 0.7920715 0.9300295 2.6844779
## [8] 1.7990998 2.6894584 0.5445647