Selecting frequent values from multiple columns from R table

Question

I have a data.table res which has data as follows:

            V1 V2 V3 V4
  1:     Day_1  4  4  4
  2:     Day_2  1  1  2
  3:     Day_3  4  5  4
  4:     Day_4  3  4  4
  5:     Day_5  3  2  3

I need to select the most frequent value from the columns V2, V3 and V4 combined. That is, I need to select result as follows:

Day_1 4
Day_2 1
Day_3 4
Day_4 4
Day_5 3

I'm not expecting any ties since the there will be always odd number of columns. Is it possible to manipulate the data.table to do this? Or should I modify it some other data type?

Thanks - V

Answer 1

I'm posting this as a data.table version of this old question until something better is offered

Mode <- function(x) {
  ux <- unique(x)  
  ux[which.max(tabulate(match(x, ux)))]
}

DT[, .(res = Mode(unlist(.SD))), by = V1]

#       V1 res
# 1: Day_1   4
# 2: Day_2   1
# 3: Day_3   4
# 4: Day_4   4
# 5: Day_5   3

Answer 2

Convert to long form and then it's trivial to do:

dt <- data.table(id=paste("Day",1:5,sep="_"),V2=c(4,1,4,3,3),V3=c(4,1,5,4,2),V4=c(4,2,4,4,3))

melt(dt, id.vars = 'id')[, .N, by = .(id, value)][, value[which.max(N)], by = id]
#      id V1
#1: Day_1  4
#2: Day_2  1
#3: Day_3  4
#4: Day_4  4
#5: Day_5  3

This is significantly faster than the other options so far, as long as number of unique (id,value) pairs is not too large.