R/transform.R

In bases: Basis Expansions for Regression Modeling

# Helper to handle rescaling
do_std <- function(x, stdize = c("scale", "box", "symbox", "none"), shift = NULL, scale = NULL) {
    stdize = rlang::arg_match(stdize, error_call=parent.frame())

    if (stdize == "scale") {
        if (is.null(shift)) shift = colMeans(x)
        if (is.null(scale)) scale = apply(x, 2, sd)
    } else if (stdize == "box") {
        rgs = apply(x, 2, range)
        if (is.null(shift)) shift = rgs[1, ]
        if (is.null(scale)) scale = rgs[2, ] - rgs[1, ]
    } else if (stdize == "symbox") {
        rgs = apply(x, 2, range)
        if (is.null(shift)) shift = 0.5*(rgs[1, ] + rgs[2, ])
        if (is.null(scale)) scale = rgs[2, ] - rgs[1, ]
    } else {
        shift = 0
        scale = 1
    }
    if (length(shift) == 1) shift = rep(shift, ncol(x))
    if (length(scale) == 1) scale = rep(scale, ncol(x))

    if (!all(shift == 0)) x = x - rep(shift, each=nrow(x))
    if (!all(scale == 1)) x = x / rep(scale, each=nrow(x))

    list(x=x, shift=shift, scale=scale)
}