mjhjmtu

I have a DataFrame, df.

n is a column denoting the number of groups in the x column.
x is a column containing the comma-separated groups.

df <- data.frame(n = c(2, 3, 2, 2), 
 x = c("a, b", "a, c, d", "c, d", "d, b"))

> df
n x
2 a, b
3 a, c, d
2 c, d
2 d, b

I would like to convert this DataFrame into a weights matrix where the row and column names are the unique values of the groups in df$x, and the elements represent the number of times each of the groups appear together in df$x.

The output should look like this:

m <- matrix(c(0, 1, 1, 1, 1, 0, 0, 1, 1, 0, 0, 2, 1, 1, 2, 0), nrow = 4, ncol = 4)
rownames(m) <- letters[1:4]; colnames(m) <- letters[1:4]

> m
 a b c d
a 0 1 1 1
b 1 0 0 1
c 1 0 0 2
d 1 1 2 0

Here's a very rough and probably pretty inefficient solution using tidyverse for wrangling and combinat to generate permutations.

library(tidyverse)
library(combinat)

df <- data.frame(n = c(2, 3, 2, 2), 
 x = c("a, b", "a, c, d", "c, d", "d, b"))

df %>% 
 ## Parse entries in x into distinct elements
 mutate(split = map(x, str_split, pattern = ', '), 
 flat = flatten(split)) %>% 
 ## Construct 2-element subsets of each set of elements
 mutate(combn = map(flat, combn, 2, simplify = FALSE)) %>% 
 unnest(combn) %>% 
 ## Construct permutations of the 2-element subsets
 mutate(perm = map(combn, permn)) %>% 
 unnest(perm) %>% 
 ## Parse the permutations into row and column indices
 mutate(row = map_chr(perm, 1), 
 col = map_chr(perm, 2)) %>% 
 count(row, col) %>% 
 ## Long to wide representation
 spread(key = col, value = nn, fill = 0) %>% 
 ## Coerce to matrix
 column_to_rownames(var = 'row') %>% 
 as.matrix()

Using Base R, you could do something like below

a = strsplit(as.character(df$x),', ')
b = unique(unlist(a))
d = unlist(sapply(a,combn,2,toString))
e = data.frame(table(factor(d,c(paste(b,b,sep=','),combn(b,2,toString)))))
f = read.table(text = do.call(paste,c(sep =',', e)),sep=',',strip.white = T)
g = xtabs(V3~V1+V2,f)
g[lower.tri(g)] = t(g)[lower.tri(g)]
g
 V2
V1 a b c d
 a 0 1 1 1
 b 1 0 0 0
 c 1 0 0 2
 d 1 0 2 0

Here is another possible approach using data.table:

#generate the combis
combis <- df[, transpose(combn(sort(strsplit(x, ", ")[[1L]]), 2L, simplify=FALSE)), 
 by=1L:df[,.N]]

#create new rows for identical letters within a pair or any other missing combi
withDiag <- out[CJ(c(V1,V2), c(V1,V2), unique=TRUE), on=.(V1, V2)]

#duplicate the above for lower triangular part of the matrix
withLowerTri <- rbindlist(list(withDiag, withDiag[,.(df, V2, V1)]))

#pivot to get weights matrix
outDT <- dcast(withLowerTri, V1 ~ V2, function(x) sum(!is.na(x)), value.var="df")

outDT output:

 V1 a b c d
1: a 0 1 1 1
2: b 1 0 0 1
3: c 1 0 0 2
4: d 1 1 2 0

If matrix output is desired, then

mat <- as.matrix(outDT[, -1L])
rownames(mat) <- unlist(outDT[,1L])

output:

 a b c d
a 0 1 1 1
b 1 0 0 1
c 1 0 0 2
d 1 1 2 0