R/utils-normalizer.R

In fitlandr: Fit Vector Fields and Potential Landscapes from Intensive Longitudinal Data

#' Normalizer for the input data
#'
#' This is an adapted version of [SparseVFC::norm_vecs()] with `na.rm = TRUE`.
#' @param x The matrix to be normalized. Each row of x represent a vector.
#' @noRd
normalize_vecs <- function(x) {
  x_noNA <- stats::na.omit(x)
  n_noNA <- nrow(x_noNA)
  xmean <- colMeans(x_noNA)
  x_noNA_centered <- x_noNA - matrix(rep(xmean, n_noNA), nrow = n_noNA, byrow = TRUE)
  xscale <- sqrt(sum(x_noNA_centered^2) / n_noNA)
  x_noNA_normalized <- x_noNA_centered / xscale

  x <- (x - matrix(rep(xmean, nrow(x)), nrow = nrow(x), byrow = TRUE)) / xscale

  return(
    structure(
      x,
      x_noNA = structure(
        x_noNA_normalized,
        mean = xmean,
        scale = xscale
      ),
      mean = xmean,
      scale = xscale
    )
  )
}

# several helpers
normalize_v <- function(v, ref) {
  (v - attr(ref, "mean")) / attr(ref, "scale")
}

normalize_x <- function(x, ref) {
  (x - attr(ref, "mean")[1]) / attr(ref, "scale")
}

normalize_y <- function(y, ref) {
  (y - attr(ref, "mean")[2]) / attr(ref, "scale")
}

denormalize_v <- function(v, ref) {
  v * attr(ref, "scale") + attr(ref, "mean")
}

scale_up <- function(v, ref) {
  v * attr(ref, "scale")
}

scale_up2 <- function(v, ref) {
  v * attr(ref, "scale")^2
}

denormalize_x <- function(x, ref) {
  x * attr(ref, "scale") + attr(ref, "mean")[1]
}

denormalize_y <- function(y, ref) {
  y * attr(ref, "scale") + attr(ref, "mean")[2]
}