quantile.quantile_pred: Quantiles from a distribution

In cmu-delphi/epipredict: Basic epidemiology forecasting methods

Quantiles from a distribution

Description

Given a hardhat::quantile_pred object, users may wish to compute additional quantile_levels that are not part of the object. This function attempts to estimate these quantities under some assumptions. Interior probabilities, those contained within existing probabilities are interpolated in a manner controled by the middle argument. Those outside existing probabilities are extrapolated under the assumption that the tails of the distribution decays exponentially. Optionally, one may constrain all quantiles to be within some support (say, ⁠[0, Inf)⁠).

Usage

## S3 method for class 'quantile_pred'
quantile(
  x,
  probs = seq(0, 1, 0.25),
  na.rm = FALSE,
  lower = -Inf,
  upper = Inf,
  middle = c("cubic", "linear"),
  ...
)

Arguments

x	numeric vector whose sample quantiles are wanted, or an object of a class for which a method has been defined (see also ‘details’). NA and NaN values are not allowed in numeric vectors unless na.rm is TRUE.
probs	numeric vector of probabilities with values in [0,1]. (Values up to ‘⁠2e-14⁠’ outside that range are accepted and moved to the nearby endpoint.)
na.rm	logical; if true, any NA and NaN's are removed from x before the quantiles are computed.
lower	Scalar. Optional lower bound.
upper	Scalar. Optional upper bound.
middle	Controls how extrapolation to "interior" probabilities is performed. "cubic" attempts to use stats::splinefun() while "linear" uses stats::approx(). The "linear" method is used as a fallback if "cubic" should fail for some reason.
...	unused

Value

a matrix with one row for each entry in x and one column for each value in probs

See Also

extrapolate_quantiles()

Examples

qp <- quantile_pred(matrix(1:8, nrow = 2, byrow = TRUE), 1:4 / 5)
quantile(qp)
quantile(qp, lower = 0)
quantile(qp, probs = 0.5)
quantile(qp, probs = 1:9 / 10)

cmu-delphi/epipredict documentation built on June 14, 2025, 2:41 a.m.