Generic Conformal Regression

Split-conformal prediction intervals using a pre-fitted fregre.lm model. Uses the model's FPCA components for fast prediction without refitting.

Usage

conformal.generic.regression(
  model,
  fdataobj,
  y,
  newdata,
  scalar.train = NULL,
  scalar.test = NULL,
  calibration.indices = NULL,
  cal.fraction = 0.25,
  alpha = 0.1,
  seed = NULL
)

Arguments

model

A fitted fregre.lm model object.

fdataobj

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

y

Response vector (training).

newdata

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

scalar.train

Optional scalar covariates for training.

scalar.test

Optional scalar covariates for test.

calibration.indices

Optional integer vector of 1-based indices into the training data to use as the calibration set. When provided, these observations should have been held out during model fitting so that calibration residuals are out-of-sample, restoring the coverage guarantee. If NULL (default), calibration indices are randomly selected from all training data (in-sample).

cal.fraction

Fraction of data for calibration (default 0.25). Ignored when calibration.indices is provided.

alpha

Miscoverage level (default 0.1).

seed

Random seed.

Value

Same as conformal.fregre.lm.

Warning

The model was trained on ALL data including the calibration subset, so calibration residuals are in-sample and systematically too small. The distribution-free coverage guarantee is broken. Use conformal.fregre.lm or cv.conformal.regression for valid coverage.

Examples

# \donttest{
fd <- fdata(matrix(rnorm(500), 50, 10), argvals = seq(0, 1, length.out = 10))
y <- rnorm(50)
model <- fregre.lm(fd, y)
cp <- conformal.generic.regression(model, fd, y, fd[1:10, ])
#> Warning: conformal.generic.regression uses the pre-fitted model without refitting. Calibration residuals are in-sample, so coverage guarantee is broken. Supply calibration.indices (held-out indices) for valid coverage, or use conformal.fregre.lm() / cv.conformal.regression() instead.
# }