join matching columns in a data.frame or data.tabl

问题:

I have the following data.frames:

a <- data.frame(id = 1:3, v1 = c('a', NA, NA), v2 = c(NA, 'b', 'c'))
b <- data.frame(id = 1:3, v1 = c(NA, 'B', 'C'), v2 = c("A", NA, NA))
> a
  id   v1   v2
1  1    a <NA>
2  2 <NA>    b
3  3 <NA>    c
> b
  id   v1   v2
1  1 <NA>    A
2  2    B <NA>
3  3    C <NA>

note: There are no ids for which v1 or v2 are defined in both tables; there is only a single unique non-NA value in each column for each id value

I would like to merge these data frames on matching values of "id':

ab <- merge(a, b, by = "id")

but I would also like to combine the two columns v1 and v2, so that the data.frame ab will look like this:

ab <- data.frame(id = 1:3, v1 = c("a", "B", "C"), v2 = c("A", "b", "c"))

> ab
  id v1 v2
1  1  a  A
2  2  B  b
3  3  C  c

instead, I get this:

> merge(a, b, by = "id")
  id v1.x v2.x v1.y v2.y
1  1    a <NA> <NA>    A
2  2 <NA>    b    B <NA>
3  3 <NA>    c    C <NA>

it would be helpful to have examples using both data.frame and data.table, so here are the data.table versions of above:

A <- data.table(a, key = 'id')
B <- data.table(b, key = 'id')
A[B]

回答1:

The type of merge you specify probably won't be possible using merge (with data frames), although saying that usually invites being proved wrong.

You also omit some details: will there always be a single unique non-NA value in each column for each id value? If so, this will work:

ab <- rbind(a,b)
> colFun <- function(x){x[which(!is.na(x))]}
> ddply(ab,.(id),function(x){colwise(colFun)(x)})
  id v1 v2
1  1  a  A
2  2  B  b
3  3  C  c

A similar strategy should work with data.tables as well:

abDT <- data.table(ab,key = "id")
> abDT[,list(colFun(v1),colFun(v2)),by = id]
     id V1 V2
[1,]  1  a  A
[2,]  2  B  b
[3,]  3  C  c

回答2:

If your data is as simple as it is above joran's answer is likely the simplest way. Here's may approach in base:

a <- data.frame(id = 1:3, v1 = c('a', NA, NA), v2 = c(NA, 'b', 'c'))
b <- data.frame(id = 1:3, v1 = c(NA, 'B', 'C'), v2 = c("A", NA, NA))

decider <- function(x, y) factor(ifelse(is.na(x), as.character(y), as.character(x)))
data.frame(mapply(a, b, FUN = decider))

If your data has different id's (some overlap and some do not, then here's a different approach:

a <- data.frame(id = c(1,2,4,5), v1 = c('a', NA, "q", NA), v2 = c(NA, 'b', 'c', "e"))
b <- data.frame(id = 1:4, v1 = c(NA, "A", "C", 'B'), v2 = c("A", NA, "D", NA))

decider <- function(x, y) factor(ifelse(is.na(x), as.character(y), as.character(x)))

DF <- data.frame(mapply(a, b, FUN = decider))
DF2 <- rbind(b[!b$id %in% DF$id , ], DF)
DF2 <- DF2[order(DF2$id), ]
rownames(DF2) <- 1:nrow(DF2)