pred_int_conformal: Prediction interval for a 'numeric' based on conformal...

In femiguez/predintma: Prediction Intervals for Meta-Analysis

Description

Conformal inference prediction interval for a numeric

Usage

pred_int_conformal(
  x,
  neval = 200,
  level = 0.95,
  point.pred = c("mean", "median"),
  method = c("quantile", "deviation", "jackknife")
)

Arguments

x	should be an object of class "numeric"
neval	number of evaluations to perform
level	coverage level with default 0.95
point.pred	point prediction either 'mean' or 'median'
method	totally experimental either 'quantile' or 'deviation'

Details

Calculate prediction interval based on conformal inference.

I wrote this function after reading this tutorial: https://cdsamii.github.io/cds-demos/conformal/conformal-tutorial.html

Value

a prediction interval

Examples

## Not run: 
## Prediciton interval for simulated data
set.seed(123)
x <- rnorm(100)
pdi.t <- pred_int_tdist(x) 
pdi.q <- pred_int_conformal(x)
pdi.d <- pred_int_conformal(x, method = "deviation")
pdi.j <- pred_int_conformal(x, method = "jackknife")
## Not really a prediction interval, but related
pdi.q <- quantile(x, probs = c(0.5, 0.025, 0.975))

## Following example from the link above
set.seed(12345)
n <- 1000
U <- rnorm(n)
hist(U)
u <- 2 ## New proposed value
Ua <- c(U,u) ## Augmented set
hist(Ua)
abline(v=u)
print(round(mean(Ua > u), 3))
nEval <- 200
u.candidate <- seq(from=min(U), to=max(U), length = nEval)
Cbounds <- c(u.candidate[match(5, 
                                 apply(as.matrix(u.candidate),1,
                                 function(x){floor(2.5+100*(mean(c(U, x) < x)))}))],
                                 rev(u.candidate)[match(95, rev(apply(as.matrix(u.candidate),1,
                                 function(x){ceiling(-2.5+100*(mean(c(U, x) < x)))})))])
print(Cbounds)
hist(U)
abline(v = Cbounds)
## Alternatively
pdi.cb <- pred_int_conformal(U)

## From Shafer and Vovk (pg. 5)
## Numners generated by Emmanuel Czuber
xn <- c(17,20,10,17,12,15,19,22,17,19,14,22,18,17,13,12,18,15,17)
pdi.xn <- pred_int_conformal(xn, neval = 500)

## End(Not run)

femiguez/predintma documentation built on July 5, 2021, 4:16 a.m.