extremeStat (version 1.5.9)

quantGPD: Fast GPD quantile estimate

Description

Fast GPD quantile estimate through L-moments

Usage

quantGPD(
  x,
  probs = c(0.8, 0.9, 0.99),
  truncate = 0,
  threshold = berryFunctions::quantileMean(x, truncate),
  addn = TRUE,
  quiet = FALSE,
  ...
)

Value

Vector with quantiles

Arguments

x

Vector with numeric values. NAs are silently ignored.

probs

Probabilities. DEFAULT: c(0.8,0.9,0.99)

truncate, threshold

Truncation proportion or threshold. DEFAULT: 0, computed See q_gpd.

addn

Logical: add element with sample size (after truncation). DEFAULT: TRUE

quiet

Should messages from this function be suppressed? DEFAULT: FALSE

...

Further arguments passed to lmomco::pargpa

Author

Berry Boessenkool, , Jun 2017

See Also

q_gpd for a comparison across R packages and methods, distLquantile to compare distributions

Examples

data(annMax)
quantile(annMax, 0.99)
quantGPD(annMax, 0.99)

if (FALSE)  # Excluded from CRAN checks to reduce checking time
data(rain, package="ismev") ;  rain <- rain[rain>0]
hist(rain, breaks=50, col=7)
tr <- seq(0,0.999, len=50)
qu <- pbapply::pbsapply(tr, quantGPD, x=rain, probs=c(0.9,0.99,0.999) ) # 30 s
plot(tr, qu[3,], ylim=range(rain), las=1, type="l")
lines(tr, qu[2,], col=2); lines(tr, qu[1,], col=4)

tr <- seq(0.88,0.999, len=50)
qu <- pbapply::pbsapply(tr, quantGPD, x=rain, probs=c(0.9,0.99,0.999) ) # 5 s
plot(tr, qu[3,], ylim=range(rain), las=1, type="l")
lines(tr, qu[2,], col=2); lines(tr, qu[1,], col=4);
tail(qu["n",])

library(microbenchmark)
data(rain, package="ismev"); rain <- rain[rain>0]
mb <- microbenchmark(quantGPD(rain[1:200], truncate=0.8, probs=0.99, addn=F),
distLquantile(rain[1:200], sel="gpa", emp=F, truncate=0.8, quiet=T, probs=0.99)[1,1]
)
boxplot(mb)
# since computing the lmoments takes most of the computational time,
# there's not much to optimize in large samples like n=2000