Combining data frame based on data in each column in dplyr

Question

Say I have some network data as shown below:

col_a <- c("A","B","C")
col_b <- c("B","A","A")
val <- c(1,3,7)
df <- data.frame(col_a, col_b, val)
df

  col_a col_b val
1     A     B   1
2     B     A   3
3     C     A   7

This could be a network and val could be the weight of the edges between the two. However, I want to add the weight between A and B and B and A to get the following:

new_col_a <- c("A", "A")
new_col_b <- c("B", "C")
new_val <- c(4,7)
want_df <- data.frame(new_col_a, new_col_b, new_val)
want_df

  new_col_a new_col_b new_val
1         A         B       4
2         A         C       7

Is there a way to do this in dplyr?

Answer 1

One dplyr possibility could be:

df %>%
 mutate_if(is.factor, as.character) %>%
 group_by(grp = paste(pmin(col_a, col_b), pmax(col_a, col_b), sep = "_")) %>%
 summarise(val = sum(val))

  grp     val
  <chr> <dbl>
1 A_B       4
2 A_C       7

Or with tidyverse, using a similar similar idea as @Sonny:

df %>%
 mutate_if(is.factor, as.character) %>%
 nest(col_a, col_b) %>%
 group_by(grp = unlist(map(data, function(x) paste(sort(x), collapse = "_")))) %>%
 summarise(val = sum(val))

If you want to also separate it into two columns (this step will also require tidyr):

df %>%
 mutate_if(is.factor, as.character) %>%
 group_by(grp = paste(pmin(col_a, col_b), pmax(col_a, col_b), sep = "_")) %>%
 summarise(val = sum(val)) %>%
 separate(grp, c("new_col_a", "new_col_b"), sep = "_")

  new_col_a new_col_b   val
  <chr>     <chr>     <dbl>
1 A         B             4
2 A         C             7

Or in the case of second possibility:

df %>%
 mutate_if(is.factor, as.character) %>%
 nest(col_a, col_b) %>%
 group_by(grp = unlist(map(data, function(x) paste(sort(x), collapse = "_")))) %>%
 summarise(val = sum(val)) %>%
 separate(grp, c("new_col_a", "new_col_b"), sep = "_")

Answer 2

You could use dplyr for this

df <- data.frame(col_a, col_b, val, stringsAsFactors = F)

library(dplyr)
library(tidyr)
df %>% 
  mutate(
    pair = purrr::pmap_chr(
      .l = list(from = col_a, to = col_b),
      .f = function(from, to) paste(sort(c(from, to)), collapse = "_")
    )
  ) %>%
  group_by(pair) %>%
  summarise(new_val = sum(val)) %>%
  separate(pair, c("new_col_a", "new_col_b"), sep = "_")
  # A tibble: 2 x 3
  new_col_a new_col_b new_val
  <chr>     <chr>       <dbl>
1 A         B               4
2 A         C               7

Similar to one of my earlier answers

Answer 3

If you make your data into a tidy, long form first, then it becomes quite a bit simpler. Convert to long, sort your column labels independent of your values, group, sum your val:

df %>%
    gather(grp,col,-val) %>%
    mutate(col=col[order(col,grp)]) %>%
    spread(grp,col) %>%
    group_by(col_a, col_b) %>%
    summarize(val = sum(val))

## A tibble: 2 x 3
## Groups:   col_a [?]
#  col_a col_b   val
#  <chr> <chr> <dbl>
#1 A     B         4
#2 A     C         7