Leave-one-out cross-validation for linear models

Computes exact leave-one-out (LOO) prediction errors by refitting the model n times, each time dropping one observation. Uses qr_downdate internally to avoid recomputing the full factorization from scratch.

Usage

lm_loo_cv(X, y, method = "qr", ...)

Arguments

X

Numeric matrix or adgeMatrix of predictors, shape [n, p]. No intercept column is added automatically.

y

Numeric vector or single-column matrix of responses, length n.

method

Character string passed to am_qr controlling the QR algorithm. Currently only "qr" is supported.

...

Additional arguments forwarded to am_qr.

Value

A named list with two elements:

residuals

Numeric vector of length n: LOO prediction errors \(y_i - \hat{y}_i^{(-i)}\).

mse

Scalar mean squared LOO error.

See also

lm_fit, many_lm

Examples

if (FALSE) { # \dontrun{
X <- adgeMatrix(matrix(rnorm(50), nrow = 10))
y <- rnorm(10)
cv <- lm_loo_cv(X, y)
cv$mse
} # }