R/utility_functions.R

In linearOrdering: Methods of Linear Ordering of Data

# Utility functions

# Normalize impacts to all positive by replacing negative influences
# with their negatives.
#
# @param decision A numeric matrix with m rows for m alternatives and n columns
#   for n criterions.
# @param impacts A character vector of '+' and '-' signs for the way that each
#   criterion influences on the alternatives.
# @return A copy of \code{decision} matrix with normalized impacts.
# @author Antoni Baum \email{antoni.baum@protonmail.com}
# @examples
# d <- matrix(rpois(12, 5), nrow = 3, ncol = 3)
# i <- c('+', '-', '+')
# .normalize_impacts(d, i)
.normalize_impacts <- function(decision, impacts) {
    if (length(impacts) != ncol(decision))
        warning("length of 'impacts' is not equal to number of columns, treating every next column as positively influencing")
    decision[, impacts == "-"] <- -decision[, impacts == "-"]
    return(decision)
}

# Calculate Euclidean distance between two vectors.
#
# @param x,y Numeric vectors.
# @return The distance between the vectors.
# @author Antoni Baum \email{antoni.baum@protonmail.com}
# @examples
# x <- c(2, 5)
# y <- c(5, 7)
# .calculate_distance(x, y)
.calculate_distance <- function(x, y) {
    sqrt(sum((x - y)^2))
}

# Normalization for TOPSIS. Intended to be used with \code{apply} function
# on a target matrix.
#
# @param decision A numeric matrix with m rows for m alternatives and n columns
#   for n criterions.
# @param weights A numeric vector with length equal to number of columns in
#   decision matrix for weights of criterions.
# @return A copy of \code{decision} matrix with TOPSIS-normalized values.
# @author Antoni Baum \email{antoni.baum@protonmail.com}
# @references Hwang, C.L.; Yoon, K. (1981). Multiple Attribute Decision
# Making: Methods and Applications. New York: Springer-Verlag.
# @examples
# d <- matrix(rpois(12, 5), nrow = 3, ncol = 3)
# w <- c(1, 1, 2)
# d <- sapply(d, .topsis_normalize, w)
.topsis_normalize <- function(decision, weights) {
    weights * decision / sqrt(sum(decision^2))
}