VIKOR function from MCDM R package is returning error

Question

I am trying to run VIKOR from MCDM R package using the following code

library(MCDM)

 d <- matrix(c(1,2,5,3000,3750,4500),nrow = 3,ncol = 2)
 w <- c(0.5,0.5)
 cb <- c('min','max')
 v <- 0.5
 VIKOR(d,w,cb,v)

But it returns me following error

Error in (Q == "NaN") || (Q == "Inf") : 'length = 3' in coercion to 'logical(1)'

How can I solve this error?

The source code for VIKOR is available here.

Answer 1

I have solved this problem by modifying the source code like

VIKOR <- function(decision, #matrix with all the alternatives
                  weights,  #vector with the numeric values of the weights
                  cb,       #vector with the "type" of the criteria (benefit = "max", cost = "min")
                  v         #value with the real number of the 'v' parameter to calculate Q
)
{
  #Checking parameters
  if(! is.matrix(decision))
    stop("'decision' must be a matrix with the values of the alternatives")
  if(missing(weights))
    stop("a vector containing n weigths, adding up to 1, should be provided")
  if(sum(weights) != 1)
    stop("The sum of 'weights' is not equal to 1")
  if(! is.character(cb))
    stop("'cb' must be a character vector with the type of the criteria")
  if(! all(cb == "max" | cb == "min"))
    stop("'cb' should contain only 'max' or 'min'")
  if(length(weights) != ncol(decision))
    stop("length of 'weights' does not match the number of the criteria")
  if(length(cb) != ncol(decision))
    stop("length of 'cb' does not match the number of the criteria")
  if(missing(v))
    stop("a value for 'v' in [0,1] should be provided")
  
  #1. Ideal solutions
  posI <- as.integer(cb == "max") * apply(decision, 2, max) +
    as.integer(cb == "min") * apply(decision, 2, min)
  negI <- as.integer(cb == "min") * apply(decision, 2, max) +
    as.integer(cb == "max") * apply(decision, 2, min)
  
  #2. S and R index
  norm =function(x,w,p,n){
    w*((p-x)/(p-n))
  }
  SAux <- apply(decision, 1, norm, weights, posI, negI)
  S <- apply(SAux, 2, sum)
  R <- apply(SAux, 2, max)
  
  
  #3. Q index
  #If v=0
  if (v==0)
    Q <- (R-min(R))/(max(R)-min(R))
  #If v=1
  else if (v==1)
    Q <- (S-min(S))/(max(S)-min(S))
  #Another case
  else
    Q <- v*(S-min(S))/(max(S)-min(S))+(1-v)*(R-min(R))/(max(R)-min(R))
  
  #4. Checking if Q is valid
  #Here is the change by taking help from https://stackoverflow.com/questions/76667355/how-to-ignore-ranking-if-the-data-conatins-nan-or-inf-in-r
  RankingQ = ifelse(is.infinite(Q) | is.na(Q), NA,
         rank(Q, ties.method = "first"))
  
  #5. Ranking the alternatives
  return(data.frame(Alternatives = 1:nrow(decision), S = S, R = R, Q = Q, 
                    Ranking = RankingQ))
  
}

Now it is working fine like

d <- matrix(c(1,2,5,3000,3750,4500),nrow = 3,ncol = 2)
w <- c(0.5,0.5)
cb <- c('min','max')
v <- 0.5
VIKOR(d,w,cb,v)

#>  Alternatives     S    R Q Ranking
#> 1            1 0.500 0.50 1       2
#> 2            2 0.375 0.25 0       1
#> 3            3 0.500 0.50 1       3

Answer 2

Where does this error come from?

This error appeared in R 4.3.0 (it was a warning since 4.2.0). From the NEWS:

Calling && or || with LHS or (if evaluated) RHS of length greater than one is now always an error, with a report of the form

'length = 4' in coercion to 'logical(1)'

In the source code you link to, the problem is this line at the end of VIKOR():

if( (Q == "NaN") || (Q == "Inf")){

The problem is that, in your case, Q returns a vector:

1 0 1

so each of the two statements returns a length-3 vector:

> print((Q == "NaN"))
[1] FALSE FALSE FALSE

> print((Q == "Inf"))
[1] FALSE FALSE FALSE

Using || with these two vectors whose length is larger than 1 throws an error.

What are your options?

1. Check that your input is correct.

2. Contact the package author. Apparently the package isn't on CRAN anymore (removed in April 2022) but you can still get the email address from the DESCRIPTION file.

3. The function in itself is not hard to extract from the package. It doesn't use any internal function or another package. If you're comfortable enough with the code and the method behind, you could try to fix it yourself.