Question

I have a dataset with 10 columns. The first column is an unique identifier. The 9 other columns are related attributes. For now, let’s just say they are integers. If needed, the data could easily be pivoted to a key-value.

Ex:

id|attr1|attr2|attr3|...
a |  2  |  5  |  7  |...
b |  3  |  1  |null |...
c |  2  |null |null |...
d |  1  |  2  |  5  |...
e |  2  |  1  |  3  |...

I’m essentially looking for the most frequent combinations of any length with at least a pair. So my output for this would be:

unq   | frequency
1,2   | 2
1,3   | 2
1,5   | 1
2,3   | 1
2,5   | 2
2,7   | 1
1,2,3 | 1
1,2,5 | 1
2,5,7 | 1

(did this manually – so hopefully there are no errors) – the order of the paring doesn’t matter. 2,5,7 = 5,2,7 = 7,5,2 etc.

Any thoughts? I am open to different tools. I have access to R, excel, sql server, mysql, etc.

Excel is preferred but not required!

Answer 1

Here is a solution in R:

Recreate the data

x <- data.frame(
    id = letters[1:5],
    attr1 = c(2,3,2,1,2),
    attr2 = c(5,1,NA,2,1),
    attr3 = c(7,NA,NA,5,3))
x

  id attr1 attr2 attr3
1  a     2     5     7
2  b     3     1    NA
3  c     2    NA    NA
4  d     1     2     5
5  e     2     1     3

Create a function to list all the combinations

make_combinations <- function(data, size){
  t1 <- apply(data[, -1], 1, function(data)unname(sort(data)))
  t2 <- lapply(t1, function(xt){if(length(xt)>=size){combn(xt, size)}})
  t3 <- sapply(t2[!is.na(t2)], 
      function(chunk){if(!is.null(chunk))apply(chunk, 2, function(x)paste(x, collapse=","))})
  t4 <- unlist(t3)
  t4
}

Create a second function to count the combinations

count_combinations <- function(data, nn=2:3){
  tmp <- unlist(lapply(nn, function(n)make_combinations(data, n)))
  sort(table(tmp), decreasing=TRUE)
}  

The results:

count_combinations(x, 2:3)


  1,2   1,3   2,5 1,2,3 1,2,5   1,5   2,3 2,5,7   2,7   5,7 
    2     2     2     1     1     1     1     1     1     1