Joining two datasets on comma-delimited column

Question

I have a large dataset that I'm coding with both human-readable and machine-readable identifiers. I'd like to type in only the human-readable codes, and use a merge in R to add the machine-readable ones. Only hitch is that I'm adding multiple identifiers into the column, separated by commas. It looks a bit like this:

df <- as.data.frame(cbind(identifier=c("a","a, b","b","b, c","c"), data=c(1,2,3,4,5)))

codebook <- as.data.frame(cbind(id=c("a","b", "c","d"),code=c('9999','8888','7777','6666')))

What I'd like to get in this end would look like this:

 answer <- as.data.frame(cbind(identifier=c("a","a, b","b","b, c","c"), code=c('9999', '9999, 8888', '8888', '8888, 7777', '7777'), data=c(1,2,3,4,5)))

I've experimented with separate() and unite() in dplyr, but I'm wondering if there's a simpler way.

Answer 1

This doesn't give your your exact output, but it may be more desirable as it is easy to work with (it is more "tidy", if you like the Wickham verbiage):

df %>%
  mutate(new_1 = gsub("(.*)[, ](.*)", "\\1", identifier),
         new_2 = gsub("(.*)[, ](.*)", "\\2", identifier)) %>%
  mutate(new_2 = ifelse(new_1 == new_2, NA, new_2)) %>%
  select(data, new_1, new_2) %>%
  melt("data") %>%
  inner_join(codebook, by = c("value" = "id"))

#   data variable value code
# 1    1    new_1     a 9999
# 2    2    new_1     a 9999
# 3    3    new_1     b 8888
# 4    4    new_1     b 8888
# 5    5    new_1     c 7777
# 6    2    new_2     b 8888
# 7    4    new_2     c 7777

Answer 2

Try separate_rows . First convert the factor columns to character. Then use separate_rows to unnest df , left join it to codebook and convert back. Note that the result has character columns.

library(dplyr)
library(tidyr)

df %>%
   mutate_all(as.character) %>%
   separate_rows(identifier) %>% 
   left_join(codebook %>% mutate_all(as.character), by = c("identifier" = "id")) %>% 
   group_by(data) %>% 
   summarize(identifier = toString(identifier), code = toString(code)) %>%
   ungroup

giving:

# A tibble: 5 x 3
  data  identifier code      
  <chr> <chr>      <chr>     
1 1     a          9999      
2 2     a, b       9999, 8888
3 3     b          8888      
4 4     b, c       8888, 7777
5 5     c          7777