Cross-Conformal (CV+) Regression

Constructs prediction intervals using K-fold cross-conformal inference. No data is wasted on a calibration split; each fold provides out-of-fold calibration residuals.

Usage

cv.conformal.regression(
  fdataobj,
  y,
  newdata,
  method = c("fregre.lm", "fregre.np"),
  scalar.train = NULL,
  scalar.test = NULL,
  ncomp = 3,
  h.func = 0,
  h.scalar = 0,
  n.folds = 5,
  alpha = 0.1,
  seed = NULL
)

Arguments

fdataobj

An object of class 'fdata' (training data).

y

Response vector (training).

newdata

An object of class 'fdata' (test data).

method

Regression method: "fregre.lm" or "fregre.np".

scalar.train

Optional scalar covariates for training.

scalar.test

Optional scalar covariates for test.

ncomp

Number of FPC components (default 3, for fregre.lm).

h.func

Functional bandwidth (default 0 = auto, for fregre.np).

h.scalar

Scalar bandwidth (default 0 = auto, for fregre.np).

n.folds

Number of folds (default 5).

alpha

Miscoverage level (default 0.1 for 90 percent intervals).

seed

Random seed.

Value

A list with components:

predictions

Point predictions for test data

lower

Lower bounds of prediction intervals

upper

Upper bounds of prediction intervals

residual.quantile

Calibration residual quantile

coverage

Empirical coverage on calibration set

Examples

# \donttest{
fd <- fdata(matrix(rnorm(500), 50, 10), argvals = seq(0, 1, length.out = 10))
y <- rnorm(50)
cp <- cv.conformal.regression(fd[1:40, ], y[1:40], fd[41:50, ])
# }