Calculate rank with ties based on more than one va

问题:

I'm trying to compute a medal table for a sports event.

My data looks like this:

test <- data.frame("ID" = c("1_1", "1_2", "1_3", "1_4","1_5","1_6"),
                   "gold"=c(10, 4, 1, 7, 7, 1),
                   "silver"=c(1, 3, 2, 19, 19, 2),
                   "bronze"=c(1, 8, 2, 0, 0, 2))

First, I want to order the data based on number of "gold", "silver", and "bronze", like this:

(test_ordered <- with(test, test[order(-gold, -silver, -bronze), ]))

Then compute the final medal rank. This is how the final rank column should like:

(test_ordered$rank<-c(1, 2, 2, 4, 5, 5))

 #    ID gold silver bronze rank
 # 1 1_1   10      1      1    1
 # 4 1_4    7     19      0    2
 # 5 1_5    7     19      0    2
 # 2 1_2    4      3      8    4
 # 3 1_3    1      2      2    5
 # 6 1_6    1      2      2    5

As ID "1_4" and "1_5" have the won the same combination of medals they'd share rank 2, e.g.

My attempts using more than two criteria with rank (also dplyr::min_ranked) failed:

with(test, rank(-gold, -silver, -bronze, ties.method = "min")) 
# (...) unused argument (-bronze)

Also interaction was not successful:

as.numeric(interaction(gl(-test$gold), gl(-test$silver), gl(-test$bronze), lex.order = TRUE))

Any ideas how to calculate rank based on multiple variables?

---

solved using henrik's idea:

as.data.frame(setDT(test)[ , rank := frank(test, -gold, -silver, -bronze, ties.method = "min")]; setorder(test, rank))

回答1:

You may use the data.table equivalent of base::rank, frank. A nice feature with frank is that it accepts, not only vectors (as in rank), but also a data.frame or a data.table as input. For these types of objects, the rank may be based on several columns.

Using your original data.frame:

test$rank <- data.table::frank(test, -gold, -silver, -bronze, ties.method = "min")

Or if you want to go all in with data.table functions:

setDT(test)[ , rank := frank(test, -gold, -silver, -bronze, ties.method = "min")]
setorder(test, rank)

回答2:

A base R solution would be:

test <- data.frame("ID"=c("1_1", "1_2", "1_3", "1_4","1_5","1_6"), 
                   "gold"=c(10,4,1,7,7,1), 
                   "silver"=c(1,3,2,19,19,2), 
                   "bronze"=c(1,8,2,0,0,2))

(test_ordered<-with(test, test[order(-gold,-silver,-bronze),]))

roll.any.greater <- function (mat) {
  mat.lead <- head(mat, -1)
  mat.lag <- tail(mat, -1)
  result <- rep(1, nrow(mat.lead) + 1)
  for (i in (2:length(result))) {
    result[i] <- ifelse(any(as.logical(abs(mat.lead[i-1, ] - mat.lag[i-1, ]))) != FALSE,
                        i, result[i-1])
  }
  return(result)
}
(want <- cbind(test_ordered,
               rank =
               roll.any.greater(test_ordered[colnames(test_ordered) %in% c("gold", "silver", "bronze")])))