Computes the Bayesian Information Criterion for a basis representation. BIC penalizes complexity more strongly than AIC for larger samples.
basis.bic(
fdataobj,
nbasis,
type = c("bspline", "fourier"),
lambda = 0,
pooled = TRUE
)The BIC value (scalar).
An fdata object.
Number of basis functions.
Basis type: "bspline" (default) or "fourier".
Smoothing/penalty parameter (default 0).
Logical. If TRUE (default), compute a single BIC across all curves. If FALSE, compute BIC for each curve and return the mean.
BIC is computed as: $$BIC = n \log(RSS/n) + \log(n) \cdot edf$$
When pooled = TRUE, the criterion uses total observations and total
effective degrees of freedom (n_curves * edf). When pooled = FALSE,
the criterion is computed for each curve separately and the mean is returned.