How to find mutual pairs in two data frames after removing duplicate rows in both using a condition in R

Question

The data I have is something like that:

RES1 <- c("A","B","A","A","B")
RES2 <- c("B","A","A","B","A")
VAL1 <-c(3,5,3,6,8)
VAL2 <- c(5,3,7,2,7)
dff <- data.frame(RES1,VAL1,RES2,VAL2)
dff
  RES1 VAL1 RES2 VAL2
  1    A    3    B    5 
  2    B    5    A    3
  3    A    3    A    7
  4    A    6    B    2
  5    B    8    A    7

I want to remove the lines where I already have the same res1-res2 pair. For example: A 3 interacts with B 5. That's the information I want. I do not care which pair is first. B 5 with A 3 or A 3 with B 5. What I want to get is the following dataframe:

output
  RES1 VAL1 RES2 VAL2
   1    A    3    B    5
   2    A    3    A    7
   3    A    6    B    2
   4    B    8    A    7

Then I want to do the same for another data frame such as :

RES3 <- c("B","B","B","A","B")
RES4 <- c("A","A","A","A","B")
VAL4 <- c(3,7,5,3,8)
VAL3 <- c(5,8,3,7,3)
df2 <- data.frame(RES3,VAL3,RES4,VAL4)

df2
  RES3 VAL3 RES4 VAL4
   1     B     5     A     3
   2     B     8     A     7
   3     B     3     A     5
   4     A     7     A     3
   5     B     3     B     8

At the end, I just want to keep mutual pairs (in my definition both pairs are the same, keeping one is essential : "A 5" - "B 3" is the same as "B 3" - "A 5". In other words, order does not matter.

Final output I desire should have the following pairs which are unique and which exist in BOTH dataframes:

mutualpairs
  RESA VALA RESB VALB
  A     3     B     5
  A     3     A     7
  B     8     A     7

Answer 1

Using the logic in my earlier comment, you could try this

df2 <- structure(list(RES3 = c("B", "B", "B", "A", "B"), VAL3 = c(5, 8, 3, 7, 3), RES4 = c("A", "A", "A", "A", "B"), VAL4 = c(3, 7, 5, 3, 8)), .Names = c("RES3", "VAL3", "RES4", "VAL4"), row.names = c(NA, -5L), class = "data.frame")
dff <- structure(list(RES1 = c("A", "B", "A", "A", "B"), VAL1 = c(3, 5, 3, 6, 8), RES2 = c("B", "A", "A", "B", "A"), VAL2 = c(5, 3, 7, 2, 7)), .Names = c("RES1", "VAL1", "RES2", "VAL2"), row.names = c(NA, -5L), class = "data.frame")

Write a function that splits a data frame into pairs of columns and sorts the interaction pairs

f <- function(data) {
  idx <- t(apply(sapply(seq(ncol(data) / 2), function(x)
    interaction(data[, 2 * x + -1:0], sep = ' ')), 1, sort))
  apply(idx, 1, paste, collapse = ' ')
}

Eg,

f(dff)
# [1] "A 3 B 5" "A 3 B 5" "A 3 A 7" "A 6 B 2" "A 7 B 8"

Use it on both data frames and find the common ones; then you could use this index to 1) subset the original data frames or 2) since this already contains the information you want, just undo what f does, eg

dff$idx <- f(dff)
df2$idx <- f(df2)

idx <- intersect(dff$idx, df2$idx)

read.table(text = idx, col.names = c('RESA','VALA','RESB','VALB'))
#   RESA VALA RESB VALB
# 1    A    3    B    5
# 2    A    3    A    7
# 3    A    7    B    8

Answer 2

Here is one solution:

library(dplyr)
df1$combined <- apply(df1, 1, function(x) paste(sort(c(paste(x[1], x[2], collapse = ','), paste(x[3], x[4], collapse = ','))), collapse = ','))
df2$combined <- apply(df2, 1, function(x) paste(sort(c(paste(x[1], x[2], collapse = ','), paste(x[3], x[4], collapse = ','))), collapse = ','))

df <- inner_join(df1 %>% group_by(combined) %>% slice(1), df2 %>% group_by(combined) %>% slice(1))
df <- df %>% ungroup() %>% select(RES1, VAL1, RES2, VAL2)

Giving you this output:

Source: local data frame [3 x 4]

    RES1  VAL1   RES2  VAL2
  (fctr) (dbl) (fctr) (dbl)
1      A     3      A     7
2      A     3      B     5
3      B     8      A     7

This approach does only one pass through the data in the above apply functions and relies on highly efficient dplyr group_by and inner_join.

The main task is to get the pairs of columns in a certain order so they can be then grouped by it. The first two lines are simply getting two pairs of columns 'appended' into a string that is ordered so values appearing in either pair count as same. Then, the group_by() is grouping all rows with same combined column and taking the first row (using slice). Joining the resulting two data frames using inner_join ensures only rows that are common between the two data frames are retained. The last line is simply selecting desired columns to keep. BTW - with respect to the message above, you may want to specify by argument in the inner_join call if you only want to join by combined column and nothing else.