R/normalizations_R.R

In TSrepr: Time Series Representations

# ArcTan normalisation to (-1,1) range ----

#' @rdname norm_atan
#' @name norm_atan
#' @title Arctangent normalisation
#'
#' @description The \code{norm_atan} normalises time series by Arctangent to max (-1,1) range.
#'
#' @return the numeric vector of normalised values
#'
#' @param x the numeric vector (time series)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{norm_z}, \link[TSrepr]{norm_min_max}}
#'
#' @examples
#' norm_atan(rnorm(50))
#'
#' @export norm_atan
norm_atan <- function(x) {

  x <- as.numeric(x)

  norm_values <- atan(x) / (pi / 2)


  return(norm_values)

}

#' @rdname denorm_atan
#' @name denorm_atan
#' @title Arctangent denormalisation
#'
#' @description The \code{denorm_atan} denormalises time series from Arctangent function.
#'
#' @return the numeric vector of denormalised values
#'
#' @param x the numeric vector (time series)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{denorm_z}, \link[TSrepr]{denorm_min_max}}
#'
#' @examples
#' denorm_atan(runif(50))
#'
#' @export denorm_atan
denorm_atan <- function(x) {

  x <- as.numeric(x)

  denorm_values <- tan(x * (pi / 2))

  return(denorm_values)

}

# Two-parameter Box-Cox normalisation -----

#' @rdname norm_boxcox
#' @name norm_boxcox
#' @title Two-parameter Box-Cox normalisation
#'
#' @description The \code{norm_boxcox} normalises time series by two-parameter Box-Cox normalisation.
#'
#' @return the numeric vector of normalised values
#'
#' @param x the numeric vector (time series)
#' @param lambda the numeric value - power transformation parameter (default is 0.1)
#' @param gamma the non-negative numeric value - parameter for holding the time series positive (offset) (default is 0)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{norm_z}, \link[TSrepr]{norm_min_max}, \link[TSrepr]{norm_atan}}
#'
#' @examples
#' norm_boxcox(runif(50))
#'
#' @export norm_boxcox
norm_boxcox <- function(x, lambda = 0.1, gamma = 0) {

  x <- as.numeric(x)

  if (gamma < 0) {

    stop("gamma must be non-negative")

  } else if (sum((x + gamma) <= 0L) > 0) {

    stop("set gamma parameter higher to be x > 0")

  }

  if (lambda == 0) {

    norm_values <- log1p(x + gamma)

  } else {

    norm_values <- (((x + gamma) ^ lambda) - 1) / lambda

  }

  return(norm_values)

}

#' @rdname denorm_boxcox
#' @name denorm_boxcox
#' @title Two-parameter Box-Cox denormalisation
#'
#' @description The \code{denorm_boxcox} denormalises time series by two-parameter Box-Cox method.
#'
#' @return the numeric vector of denormalised values
#'
#' @param x the numeric vector (time series) to be denormalised
#' @param lambda the numeric value - power transformation parameter (default is 0.1)
#' @param gamma the non-negative numeric value - parameter for holding the time series positive (offset) (default is 0)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{denorm_z}, \link[TSrepr]{denorm_min_max}, \link[TSrepr]{denorm_atan}}
#'
#' @examples
#' denorm_boxcox(runif(50))
#'
#' @export denorm_boxcox
denorm_boxcox <- function(x, lambda = 0.1, gamma = 0) {

  x <- as.numeric(x)

  if (gamma < 0) {

    stop("gamma must be non-negative")

  }

  if (lambda == 0) {

    denorm_values <- expm1(x) - gamma

  } else {

    denorm_values <- (((x*lambda) + 1)^(1/lambda)) - gamma

  }

  return(denorm_values)

}

# Yeo-Johnson normalisation -----

#' @rdname norm_yj
#' @name norm_yj
#' @title Yeo-Johnson normalisation
#'
#' @description The \code{norm_yj} normalises time series by Yeo-Johnson normalisation.
#'
#' @return the numeric vector of normalised values
#'
#' @param x the numeric vector (time series)
#' @param lambda the numeric value - power transformation parameter (default is 0.1)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{norm_z}, \link[TSrepr]{norm_min_max}, \link[TSrepr]{norm_boxcox}}
#'
#' @examples
#' norm_yj(runif(50))
#'
#' @export norm_yj
norm_yj <- function(x, lambda = 0.1) {

  x <- as.numeric(x)

  yj <- function(x, lambda) {

    if (lambda == 0L & x >= 0L) {

      norm_value <- log1p(x + 1L)

    } else if (lambda != 0L & x >= 0L) {

      norm_value <- (((x + 1L) ^ lambda) - 1L) / lambda

    } else if (lambda != 2L & x < 0L) {

      norm_value <- -((-x + 1L) ^ (2L - lambda) - 1L) / (2L - lambda)

    } else {

      norm_value <- -log1p(-x + 1L)

    }

    return(norm_value)

  }

  norm_values <- sapply(x, function(i) yj(i, lambda))

  return(norm_values)

}

#' @rdname denorm_yj
#' @name denorm_yj
#' @title Yeo-Johnson denormalisation
#'
#' @description The \code{denorm_yj} denormalises time series by Yeo-Johnson method
#'
#' @return the numeric vector of denormalised values
#'
#' @param x the numeric vector (time series) to be denormalised
#' @param lambda the numeric value - power transformation parameter (default is 0.1)
#'
#' @author Peter Laurinec, <tsreprpackage@gmail.com>
#'
#' @seealso \code{\link[TSrepr]{denorm_z}, \link[TSrepr]{denorm_min_max}, \link[TSrepr]{denorm_boxcox}}
#'
#' @examples
#' denorm_yj(runif(50))
#'
#' @export denorm_yj
denorm_yj <- function(x, lambda = 0.1) {

  x <- as.numeric(x)

  yj <- function(x, lambda) {

    if (lambda == 0L & x >= 0L) {

      denorm_value <- expm1(x) - 1L

    } else if (lambda != 0L & x >= 0L) {

      denorm_value <- (((x*lambda) + 1)^(1/lambda)) - 1

    } else if (lambda != 2L & x < 0L) {

      denorm_value <- -((-x*(2-lambda) + 1) ^ (1/(2-lambda)) - 1)

    } else {

      denorm_value <- -expm1(-x) + 1L

    }

    return(denorm_value)

  }

  denorm_values <- sapply(x, function(i) yj(i, lambda))

  return(denorm_values)

}