argus (version 0.1.1)

rargus: The Argus distribution

Description

Density, distribution function, quantile function and random generation for the Argus distribution with parameter chi. rargus() uses very efficient random variate generation methods. Inversion or the Ratio of Uniforms method can be selected.

Usage

rargus(n=length(chi) , chi, method = c("inversion","RoU") )
dargus(x, chi, log = FALSE)
pargus(x, chi, lower = TRUE, log.p = FALSE)
qargus(p, chi, lower = TRUE, log.p = FALSE)

Value

rargus creates a random sample of size n.

dargus gives the density, dargus give the CDF and qargus gives the quantile function. The length of the result for these 3 functions is the maximum of the lengths of the numerical arguments that are recycled to the length of the result. Only the first elements of the logical arguments are used.

Arguments

n

Number of observations

chi

parameter vector chi>0

method

random variate generation method to be used (details are below)

x

x-value for the density or CDF

log

If TRUE the logarithm of the density is returned.

lower

If FALSE 1-CDF or its inverse is returned

log.p

log(p) the logarithm of the probability is used or returned

p

probability of the quantile

Author

Wolfgang Hörmann and Christoph Baumgarten

Details

The Argus distribution has the density (pdf) which is proportional to $$f(x) = x \sqrt{1-x^2}\exp(-0.5\chi^2(1-x^2)), \mbox{for} 0 \leq x \leq 1 \mbox{and} \chi > 0$$

The generators used for rargus() are very efficient also for the varying parameter case, i.e. when the vector chi has length n.

method = "inversion": transforms the uniform variate into an argus variate using a close to exact approximation of the inverse CDF. It is also well suited for the varying parameter case as the theorem that an argus variate can be represented as a transformed truncated Gamma(1.5) variate allows to express the inverse CDF of the argus distribution as a simple transform of the inverse CDF of the Gamma(1.5) distribution. The use of the pinv.new() function of the Runuran package makes that evaluation very fast. When loading the argus-package pinv.new() calculates and stores the required tables.

method = "RoU": uses the Ratio of Uniforms method which requires 2 or more uniform variates to generate one argus variate. It is also a bit slower than the inversion method and is added here mainly for the case that the Runuran-package is not available.

References

Christoph Baumgarten: Random Variate Generation by Fast Numerical Inversion in the Varying Parameter Case.

Examples

library(argus)
# pdf, cdf and quantile function
dargus((0:10)/10,.6)
qargus(0.9,2)
pargus(qargus(0.9,2),2)

y<-rargus(n=1.e5,chi=0.3,method="inversion")
hist(y,breaks=100,main="Argus pdf with chi=0.3",freq=FALSE)
lines(xv<-seq(0,1,0.001),dargus(xv,chi=0.3),lwd=0.2,col=2)

n=1.e5
chiv <- 1. +runif(n)*5 # chi-values in (1,5)
system.time({set.seed(123);y<-rargus(chi=chiv)})
set.seed(123);u<- runif(n)
mean(abs(pargus(y,chiv)-u))

# random variate generation with the inversion method
y<-rargus(n=1.e5,chi=0.3,method="inversion")
hist(y,breaks=100,main="Argus pdf with chi=0.3",freq=FALSE)
lines(xv<-seq(0,1,0.001),dargus(xv,chi=0.3),lwd=0.2,col=2)
# using Ratio of Uniforms
y<-rargus(n=1.e5,chi=3,method="RoU")
hist(y,breaks=100,main="Argus pdf with chi=3",freq=FALSE)
lines(xv<-seq(0,1,0.001),dargus(xv,chi=3),lwd=2,col=2)
# generating for different chi values
y<-rargus(n=100,chi=runif(100)*2,method="RoU")