Merge two R dataframes by at least one merge ID across columns - r

Question

I have a joining problem that I'm struggling with in that the join IDs I want to use for separate dataframes are spread out across three possible ID columns. I'd like to be able to join if at least one join ID matches. I know the _join and merge functions accept a vector of column names but is it possible to make this work conditionally?
For example, if I have the following two data frames:
df_A <- data.frame(dta = c("FOO", "BAR", "GOO"),
id1 = c("abc", "", "bcd"),
id2 = c("", "", "xyz"),
id3 = c("def", "fgh", ""), stringsAsFactors = F)
df_B <- data.frame(dta = c("FUU", "PAR", "KOO"),
id1 = c("abc", "", ""),
id2 = c("", "xyz", "zzz"),
id3 = c("", "", ""), stringsAsFactors = F)
> df_A
dta id1 id2 id3
1 FOO abc def
2 BAR fgh
3 GOO bcd xyz
> df_B
dta id1 id2 id3
1 FUU abc
2 PAR xyz
3 KOO zzz
I hope to end up with something like this:
dta.x dta.y id1 id2 id3
1 FOO FUU abc "" def [matched on id1]
2 BAR "" "" "" fgh [unmatched]
3 GOO PAR bcd xyz "" [matched on id2]
4 KOO "" "" zzz "" [unmatched]
So that unmatched dta1 and dta1 variables are retained but where there is a match (row 1 + 3 above) both dta1 and dta2 are joined in the new table. I have a sense that neither _join, merge, or match will work as is and that I'd need to write a function but I'm not sure where to start. Any help or ideas appreciated. Thank you

Answer 1

Basically, what you want to do is join by corresponding IDs, what you can do is to convert the original id columns to id_column and id_value, because you don't want to join with "", do I dropped it.
library(tidyverse)
df_A_long <- df_A %>%
pivot_longer(
cols = -dta,
names_to = "id_column",
values_to = "id_value"
) %>%
dplyr::filter(id_value != "")
df_B_long <- df_B %>%
pivot_longer(
cols = -dta,
names_to = "id_column",
values_to = "id_value"
) %>%
dplyr::filter(id_value != "")
We always use id_column and id_value to join A & B.
> df_B_long
# A tibble: 3 x 3
dta id_column id_value
<chr> <chr> <chr>
1 FUU id1 abc
2 PAR id2 xyz
3 KOO id2 zzz
The joining part is clear, but to create your desired output, we need to do some data wrangling to make it look identical.
df_joined <- df_A_long %>%
# join using id_column and id_value
full_join(df_B_long, by = c("id_column","id_value"),suffix = c("1","2")) %>%
# pivot back to long format
pivot_wider(
id_cols = c(dta1,dta2),
names_from = id_column,
values_from = id_value
) %>%
# if dta1 is missing, then in the same row, move value from dta2 to dta1
mutate(
dta1_has_value = !is.na(dta1), # helper column
dta1 = ifelse(dta1_has_value,dta1,dta2),
dta2 = ifelse(!dta1_has_value & !is.na(dta2),NA,dta2)
) %>%
select(-dta1_has_value) %>%
group_by(dta1) %>%
# condense multiple rows into one row
summarise_all(
~ifelse(all(is.na(.x)),"",.x[!is.na(.x)])
) %>%
# reorder columns
{
.[sort(colnames(df_joined))]
}
Result:
> df_joined
# A tibble: 4 x 5
dta1 dta2 id1 id2 id3
<chr> <chr> <chr> <chr> <chr>
1 BAR "" "" "" fgh
2 FOO FUU abc "" def
3 GOO PAR bcd xyz ""
4 KOO "" "" zzz ""

Answer 2

library(sqldf)
one <-
sqldf('
select a.*
, b.dta as dta_b
from df_A a
left join df_B b
on a.id1 <> ""
and (
a.id1 = b.id1
or a.id2 = b.id2)
')
two <-
sqldf('
select b.*
from df_B b
left join one
on b.dta = one.dta
or b.dta = one.dta_b
where one.dta is null
')
dplyr::bind_rows(one, two)
# dta id1 id2 id3 dta_b
# 1 FOO abc def FUU
# 2 BAR fgh <NA>
# 3 GOO bcd xyz PAR
# 4 KOO zzz <NA>